@caupulican/pi-agent-core 0.81.5 → 0.81.6

package/dist/compaction/branch-summarization.d.ts

@@ -6,7 +6,7 @@
  */
 import type { Model } from "@caupulican/pi-ai";
 import type { ReadonlySessionManager, SessionEntry } from "../session/session-manager.ts";
-import type { AgentMessage } from "../types.ts";
+import type { AgentMessage, StreamFn } from "../types.ts";
 import { type FileOperations } from "./utils.ts";
 export interface BranchSummaryResult {
     summary?: string;
@@ -50,6 +50,8 @@ export interface GenerateBranchSummaryOptions {
     replaceInstructions?: boolean;
     /** Tokens reserved for prompt + LLM response (default 16384) */
     reserveTokens?: number;
+    /** Optional wrapped stream function (watchdog/retry-capable host path). */
+    streamFn?: StreamFn;
 }
 /**
  * Collect entries that should be summarized when navigating from one position to another.
@@ -78,11 +80,5 @@ export declare function collectEntriesForBranchSummary(session: ReadonlySessionM
  * @param tokenBudget - Maximum tokens to include (0 = no limit)
  */
 export declare function prepareBranchEntries(entries: SessionEntry[], tokenBudget?: number): BranchPreparation;
-/**
- * Generate a summary of abandoned branch entries.
- *
- * @param entries - Session entries to summarize (chronological order)
- * @param options - Generation options
- */
 export declare function generateBranchSummary(entries: SessionEntry[], options: GenerateBranchSummaryOptions): Promise<BranchSummaryResult>;
 //# sourceMappingURL=branch-summarization.d.ts.map

package/dist/compaction/branch-summarization.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"branch-summarization.d.ts","sourceRoot":"","sources":["../../src/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAQ/C,OAAO,KAAK,EAAE,sBAAsB,EAAE,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAC1F,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAEhD,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AAMpB,MAAM,WAAW,mBAAmB;IACnC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;CACf;AAED,qEAAqE;AACrE,MAAM,WAAW,oBAAoB;IACpC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjD,MAAM,WAAW,iBAAiB;IACjC,mEAAmE;IACnE,QAAQ,EAAE,YAAY,EAAE,CAAC;IACzB,gDAAgD;IAChD,OAAO,EAAE,cAAc,CAAC;IACxB,yCAAyC;IACzC,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,oBAAoB;IACpC,mDAAmD;IACnD,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,2DAA2D;IAC3D,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;CAChC;AAED,MAAM,WAAW,4BAA4B;IAC5C,qCAAqC;IACrC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;IAClB,4BAA4B;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,oCAAoC;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,oCAAoC;IACpC,MAAM,EAAE,WAAW,CAAC;IACpB,qDAAqD;IACrD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,wFAAwF;IACxF,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAMD;;;;;;;;;;;GAWG;AACH,wBAAgB,8BAA8B,CAC7C,OAAO,EAAE,sBAAsB,EAC/B,SAAS,EAAE,MAAM,GAAG,IAAI,EACxB,QAAQ,EAAE,MAAM,GACd,oBAAoB,CAkCtB;AAoCD;;;;;;;;;;;;GAYG;AACH,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,WAAW,GAAE,MAAU,GAAG,iBAAiB,CAoDxG;AAwCD;;;;;GAKG;AACH,wBAAsB,qBAAqB,CAC1C,OAAO,EAAE,YAAY,EAAE,EACvB,OAAO,EAAE,4BAA4B,GACnC,OAAO,CAAC,mBAAmB,CAAC,CAqE9B","sourcesContent":["/**\n * Branch summarization for tree navigation.\n *\n * When navigating to a different point in the session tree, this generates\n * a summary of the branch being left so context isn't lost.\n */\n\nimport type { Model } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage } from \"../types.ts\";\nimport { estimateTokens } from \"./compaction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface BranchSummaryResult {\n\tsummary?: string;\n\treadFiles?: string[];\n\tmodifiedFiles?: string[];\n\taborted?: boolean;\n\terror?: string;\n}\n\n/** Details stored in BranchSummaryEntry.details for file tracking */\nexport interface BranchSummaryDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\nexport type { FileOperations } from \"./utils.ts\";\n\nexport interface BranchPreparation {\n\t/** Messages extracted for summarization, in chronological order */\n\tmessages: AgentMessage[];\n\t/** File operations extracted from tool calls */\n\tfileOps: FileOperations;\n\t/** Total estimated tokens in messages */\n\ttotalTokens: number;\n}\n\nexport interface CollectEntriesResult {\n\t/** Entries to summarize, in chronological order */\n\tentries: SessionEntry[];\n\t/** Common ancestor between old and new position, if any */\n\tcommonAncestorId: string | null;\n}\n\nexport interface GenerateBranchSummaryOptions {\n\t/** Model to use for summarization */\n\tmodel: Model<any>;\n\t/** API key for the model */\n\tapiKey: string;\n\t/** Request headers for the model */\n\theaders?: Record<string, string>;\n\t/** Abort signal for cancellation */\n\tsignal: AbortSignal;\n\t/** Optional custom instructions for summarization */\n\tcustomInstructions?: string;\n\t/** If true, customInstructions replaces the default prompt instead of being appended */\n\treplaceInstructions?: boolean;\n\t/** Tokens reserved for prompt + LLM response (default 16384) */\n\treserveTokens?: number;\n}\n\n// ============================================================================\n// Entry Collection\n// ============================================================================\n\n/**\n * Collect entries that should be summarized when navigating from one position to another.\n *\n * Walks from oldLeafId back to the common ancestor with targetId, collecting entries\n * along the way. Does NOT stop at compaction boundaries - those are included and their\n * summaries become context.\n *\n * @param session - Session manager (read-only access)\n * @param oldLeafId - Current position (where we're navigating from)\n * @param targetId - Target position (where we're navigating to)\n * @returns Entries to summarize and the common ancestor\n */\nexport function collectEntriesForBranchSummary(\n\tsession: ReadonlySessionManager,\n\toldLeafId: string | null,\n\ttargetId: string,\n): CollectEntriesResult {\n\t// If no old position, nothing to summarize\n\tif (!oldLeafId) {\n\t\treturn { entries: [], commonAncestorId: null };\n\t}\n\n\t// Find common ancestor (deepest node that's on both paths)\n\tconst oldPath = new Set(session.getBranch(oldLeafId).map((e) => e.id));\n\tconst targetPath = session.getBranch(targetId);\n\n\t// targetPath is root-first, so iterate backwards to find deepest common ancestor\n\tlet commonAncestorId: string | null = null;\n\tfor (let i = targetPath.length - 1; i >= 0; i--) {\n\t\tif (oldPath.has(targetPath[i].id)) {\n\t\t\tcommonAncestorId = targetPath[i].id;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Collect entries from old leaf back to common ancestor\n\tconst entries: SessionEntry[] = [];\n\tlet current: string | null = oldLeafId;\n\n\twhile (current && current !== commonAncestorId) {\n\t\tconst entry = session.getEntry(current);\n\t\tif (!entry) break;\n\t\tentries.push(entry);\n\t\tcurrent = entry.parentId;\n\t}\n\n\t// Reverse to get chronological order\n\tentries.reverse();\n\n\treturn { entries, commonAncestorId };\n}\n\n// ============================================================================\n// Entry to Message Conversion\n// ============================================================================\n\n/**\n * Extract AgentMessage from a session entry.\n * Similar to getMessageFromEntry in compaction.ts but also handles compaction entries.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tswitch (entry.type) {\n\t\tcase \"message\":\n\t\t\t// Skip tool results - context is in assistant's tool call\n\t\t\tif (entry.message.role === \"toolResult\") return undefined;\n\t\t\treturn entry.message;\n\n\t\tcase \"custom_message\":\n\t\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\n\t\tcase \"branch_summary\":\n\t\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\n\t\tcase \"compaction\":\n\t\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\n\t\t// These don't contribute to conversation content\n\t\tcase \"thinking_level_change\":\n\t\tcase \"model_change\":\n\t\tcase \"custom\":\n\t\tcase \"label\":\n\t\tcase \"session_info\":\n\t\t\treturn undefined;\n\t}\n}\n\n/**\n * Prepare entries for summarization with token budget.\n *\n * Walks entries from NEWEST to OLDEST, adding messages until we hit the token budget.\n * This ensures we keep the most recent context when the branch is too long.\n *\n * Also collects file operations from:\n * - Tool calls in assistant messages\n * - Existing branch_summary entries' details (for cumulative tracking)\n *\n * @param entries - Entries in chronological order\n * @param tokenBudget - Maximum tokens to include (0 = no limit)\n */\nexport function prepareBranchEntries(entries: SessionEntry[], tokenBudget: number = 0): BranchPreparation {\n\tconst messages: AgentMessage[] = [];\n\tconst fileOps = createFileOps();\n\tlet totalTokens = 0;\n\n\t// First pass: collect file ops from ALL entries (even if they don't fit in token budget)\n\t// This ensures we capture cumulative file tracking from nested branch summaries\n\t// Only extract from pi-generated summaries (fromHook !== true), not extension-generated ones\n\tfor (const entry of entries) {\n\t\tif (entry.type === \"branch_summary\" && !entry.fromHook && entry.details) {\n\t\t\tconst details = entry.details as BranchSummaryDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\t// Modified files go into both edited and written for proper deduplication\n\t\t\t\tfor (const f of details.modifiedFiles) {\n\t\t\t\t\tfileOps.edited.add(f);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Second pass: walk from newest to oldest, adding messages until token budget\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tconst message = getMessageFromEntry(entry);\n\t\tif (!message) continue;\n\n\t\t// Extract file ops from assistant messages (tool calls)\n\t\textractFileOpsFromMessage(message, fileOps);\n\n\t\tconst tokens = estimateTokens(message);\n\n\t\t// Check budget before adding\n\t\tif (tokenBudget > 0 && totalTokens + tokens > tokenBudget) {\n\t\t\t// If this is a summary entry, try to fit it anyway as it's important context\n\t\t\tif (entry.type === \"compaction\" || entry.type === \"branch_summary\") {\n\t\t\t\tif (totalTokens < tokenBudget * 0.9) {\n\t\t\t\t\tmessages.unshift(message);\n\t\t\t\t\ttotalTokens += tokens;\n\t\t\t\t}\n\t\t\t}\n\t\t\t// Stop - we've hit the budget\n\t\t\tbreak;\n\t\t}\n\n\t\tmessages.unshift(message);\n\t\ttotalTokens += tokens;\n\t}\n\n\treturn { messages, fileOps, totalTokens };\n}\n\n// ============================================================================\n// Summary Generation\n// ============================================================================\n\nconst BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.\nSummary of that exploration:\n\n`;\n\nconst BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.\n\nUse this EXACT format:\n\n## Goal\n[What was the user trying to accomplish in this branch?]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Work that was started but not finished]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [What should happen next to continue this work]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\n/**\n * Generate a summary of abandoned branch entries.\n *\n * @param entries - Session entries to summarize (chronological order)\n * @param options - Generation options\n */\nexport async function generateBranchSummary(\n\tentries: SessionEntry[],\n\toptions: GenerateBranchSummaryOptions,\n): Promise<BranchSummaryResult> {\n\tconst { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;\n\n\t// Token budget = context window minus reserved space for prompt + response\n\tconst contextWindow = model.contextWindow || 128000;\n\tconst tokenBudget = contextWindow - reserveTokens;\n\n\tconst { messages, fileOps } = prepareBranchEntries(entries, tokenBudget);\n\n\tif (messages.length === 0) {\n\t\treturn { summary: \"No content to summarize\" };\n\t}\n\n\t// Transform to LLM-compatible messages, then serialize to text\n\t// Serialization prevents the model from treating it as a conversation to continue\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build prompt\n\tlet instructions: string;\n\tif (replaceInstructions && customInstructions) {\n\t\tinstructions = customInstructions;\n\t} else if (customInstructions) {\n\t\tinstructions = `${BRANCH_SUMMARY_PROMPT}\\n\\nAdditional focus: ${customInstructions}`;\n\t} else {\n\t\tinstructions = BRANCH_SUMMARY_PROMPT;\n\t}\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${instructions}`;\n\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\t// Call LLM for summarization\n\tconst response = await completeSimple(\n\t\tmodel,\n\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\t{ apiKey, headers, signal, maxTokens: 2048 },\n\t);\n\n\t// Check if aborted or errored\n\tif (response.stopReason === \"aborted\") {\n\t\treturn { aborted: true };\n\t}\n\tif (response.stopReason === \"error\") {\n\t\treturn { error: response.errorMessage || \"Summarization failed\" };\n\t}\n\n\tlet summary = response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n\n\t// Prepend preamble to provide context about the branch summary\n\tsummary = BRANCH_SUMMARY_PREAMBLE + summary;\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\treturn {\n\t\tsummary: summary || \"No summary generated\",\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}
+{"version":3,"file":"branch-summarization.d.ts","sourceRoot":"","sources":["../../src/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAA6B,KAAK,EAAuB,MAAM,mBAAmB,CAAC;AAS/F,OAAO,KAAK,EAAE,sBAAsB,EAAE,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAC1F,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAE1D,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AAMpB,MAAM,WAAW,mBAAmB;IACnC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;CACf;AAED,qEAAqE;AACrE,MAAM,WAAW,oBAAoB;IACpC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAEjD,MAAM,WAAW,iBAAiB;IACjC,mEAAmE;IACnE,QAAQ,EAAE,YAAY,EAAE,CAAC;IACzB,gDAAgD;IAChD,OAAO,EAAE,cAAc,CAAC;IACxB,yCAAyC;IACzC,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,oBAAoB;IACpC,mDAAmD;IACnD,OAAO,EAAE,YAAY,EAAE,CAAC;IACxB,2DAA2D;IAC3D,gBAAgB,EAAE,MAAM,GAAG,IAAI,CAAC;CAChC;AAED,MAAM,WAAW,4BAA4B;IAC5C,qCAAqC;IACrC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC;IAClB,4BAA4B;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,oCAAoC;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,oCAAoC;IACpC,MAAM,EAAE,WAAW,CAAC;IACpB,qDAAqD;IACrD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,wFAAwF;IACxF,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B,gEAAgE;IAChE,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,2EAA2E;IAC3E,QAAQ,CAAC,EAAE,QAAQ,CAAC;CACpB;AAMD;;;;;;;;;;;GAWG;AACH,wBAAgB,8BAA8B,CAC7C,OAAO,EAAE,sBAAsB,EAC/B,SAAS,EAAE,MAAM,GAAG,IAAI,EACxB,QAAQ,EAAE,MAAM,GACd,oBAAoB,CAkCtB;AAoCD;;;;;;;;;;;;GAYG;AACH,wBAAgB,oBAAoB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,WAAW,GAAE,MAAU,GAAG,iBAAiB,CAoDxG;AAyDD,wBAAsB,qBAAqB,CAC1C,OAAO,EAAE,YAAY,EAAE,EACvB,OAAO,EAAE,4BAA4B,GACnC,OAAO,CAAC,mBAAmB,CAAC,CAuF9B","sourcesContent":["/**\n * Branch summarization for tree navigation.\n *\n * When navigating to a different point in the session tree, this generates\n * a summary of the branch being left so context isn't lost.\n */\n\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport { classifyFailure } from \"../reliability/classifier.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage, StreamFn } from \"../types.ts\";\nimport { estimateTokens } from \"./compaction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface BranchSummaryResult {\n\tsummary?: string;\n\treadFiles?: string[];\n\tmodifiedFiles?: string[];\n\taborted?: boolean;\n\terror?: string;\n}\n\n/** Details stored in BranchSummaryEntry.details for file tracking */\nexport interface BranchSummaryDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\nexport type { FileOperations } from \"./utils.ts\";\n\nexport interface BranchPreparation {\n\t/** Messages extracted for summarization, in chronological order */\n\tmessages: AgentMessage[];\n\t/** File operations extracted from tool calls */\n\tfileOps: FileOperations;\n\t/** Total estimated tokens in messages */\n\ttotalTokens: number;\n}\n\nexport interface CollectEntriesResult {\n\t/** Entries to summarize, in chronological order */\n\tentries: SessionEntry[];\n\t/** Common ancestor between old and new position, if any */\n\tcommonAncestorId: string | null;\n}\n\nexport interface GenerateBranchSummaryOptions {\n\t/** Model to use for summarization */\n\tmodel: Model<any>;\n\t/** API key for the model */\n\tapiKey: string;\n\t/** Request headers for the model */\n\theaders?: Record<string, string>;\n\t/** Abort signal for cancellation */\n\tsignal: AbortSignal;\n\t/** Optional custom instructions for summarization */\n\tcustomInstructions?: string;\n\t/** If true, customInstructions replaces the default prompt instead of being appended */\n\treplaceInstructions?: boolean;\n\t/** Tokens reserved for prompt + LLM response (default 16384) */\n\treserveTokens?: number;\n\t/** Optional wrapped stream function (watchdog/retry-capable host path). */\n\tstreamFn?: StreamFn;\n}\n\n// ============================================================================\n// Entry Collection\n// ============================================================================\n\n/**\n * Collect entries that should be summarized when navigating from one position to another.\n *\n * Walks from oldLeafId back to the common ancestor with targetId, collecting entries\n * along the way. Does NOT stop at compaction boundaries - those are included and their\n * summaries become context.\n *\n * @param session - Session manager (read-only access)\n * @param oldLeafId - Current position (where we're navigating from)\n * @param targetId - Target position (where we're navigating to)\n * @returns Entries to summarize and the common ancestor\n */\nexport function collectEntriesForBranchSummary(\n\tsession: ReadonlySessionManager,\n\toldLeafId: string | null,\n\ttargetId: string,\n): CollectEntriesResult {\n\t// If no old position, nothing to summarize\n\tif (!oldLeafId) {\n\t\treturn { entries: [], commonAncestorId: null };\n\t}\n\n\t// Find common ancestor (deepest node that's on both paths)\n\tconst oldPath = new Set(session.getBranch(oldLeafId).map((e) => e.id));\n\tconst targetPath = session.getBranch(targetId);\n\n\t// targetPath is root-first, so iterate backwards to find deepest common ancestor\n\tlet commonAncestorId: string | null = null;\n\tfor (let i = targetPath.length - 1; i >= 0; i--) {\n\t\tif (oldPath.has(targetPath[i].id)) {\n\t\t\tcommonAncestorId = targetPath[i].id;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Collect entries from old leaf back to common ancestor\n\tconst entries: SessionEntry[] = [];\n\tlet current: string | null = oldLeafId;\n\n\twhile (current && current !== commonAncestorId) {\n\t\tconst entry = session.getEntry(current);\n\t\tif (!entry) break;\n\t\tentries.push(entry);\n\t\tcurrent = entry.parentId;\n\t}\n\n\t// Reverse to get chronological order\n\tentries.reverse();\n\n\treturn { entries, commonAncestorId };\n}\n\n// ============================================================================\n// Entry to Message Conversion\n// ============================================================================\n\n/**\n * Extract AgentMessage from a session entry.\n * Similar to getMessageFromEntry in compaction.ts but also handles compaction entries.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tswitch (entry.type) {\n\t\tcase \"message\":\n\t\t\t// Skip tool results - context is in assistant's tool call\n\t\t\tif (entry.message.role === \"toolResult\") return undefined;\n\t\t\treturn entry.message;\n\n\t\tcase \"custom_message\":\n\t\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\n\t\tcase \"branch_summary\":\n\t\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\n\t\tcase \"compaction\":\n\t\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\n\t\t// These don't contribute to conversation content\n\t\tcase \"thinking_level_change\":\n\t\tcase \"model_change\":\n\t\tcase \"custom\":\n\t\tcase \"label\":\n\t\tcase \"session_info\":\n\t\t\treturn undefined;\n\t}\n}\n\n/**\n * Prepare entries for summarization with token budget.\n *\n * Walks entries from NEWEST to OLDEST, adding messages until we hit the token budget.\n * This ensures we keep the most recent context when the branch is too long.\n *\n * Also collects file operations from:\n * - Tool calls in assistant messages\n * - Existing branch_summary entries' details (for cumulative tracking)\n *\n * @param entries - Entries in chronological order\n * @param tokenBudget - Maximum tokens to include (0 = no limit)\n */\nexport function prepareBranchEntries(entries: SessionEntry[], tokenBudget: number = 0): BranchPreparation {\n\tconst messages: AgentMessage[] = [];\n\tconst fileOps = createFileOps();\n\tlet totalTokens = 0;\n\n\t// First pass: collect file ops from ALL entries (even if they don't fit in token budget)\n\t// This ensures we capture cumulative file tracking from nested branch summaries\n\t// Only extract from pi-generated summaries (fromHook !== true), not extension-generated ones\n\tfor (const entry of entries) {\n\t\tif (entry.type === \"branch_summary\" && !entry.fromHook && entry.details) {\n\t\t\tconst details = entry.details as BranchSummaryDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\t// Modified files go into both edited and written for proper deduplication\n\t\t\t\tfor (const f of details.modifiedFiles) {\n\t\t\t\t\tfileOps.edited.add(f);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Second pass: walk from newest to oldest, adding messages until token budget\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tconst message = getMessageFromEntry(entry);\n\t\tif (!message) continue;\n\n\t\t// Extract file ops from assistant messages (tool calls)\n\t\textractFileOpsFromMessage(message, fileOps);\n\n\t\tconst tokens = estimateTokens(message);\n\n\t\t// Check budget before adding\n\t\tif (tokenBudget > 0 && totalTokens + tokens > tokenBudget) {\n\t\t\t// If this is a summary entry, try to fit it anyway as it's important context\n\t\t\tif (entry.type === \"compaction\" || entry.type === \"branch_summary\") {\n\t\t\t\tif (totalTokens < tokenBudget * 0.9) {\n\t\t\t\t\tmessages.unshift(message);\n\t\t\t\t\ttotalTokens += tokens;\n\t\t\t\t}\n\t\t\t}\n\t\t\t// Stop - we've hit the budget\n\t\t\tbreak;\n\t\t}\n\n\t\tmessages.unshift(message);\n\t\ttotalTokens += tokens;\n\t}\n\n\treturn { messages, fileOps, totalTokens };\n}\n\n// ============================================================================\n// Summary Generation\n// ============================================================================\n\nconst BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.\nSummary of that exploration:\n\n`;\n\nconst BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.\n\nUse this EXACT format:\n\n## Goal\n[What was the user trying to accomplish in this branch?]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Work that was started but not finished]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [What should happen next to continue this work]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\n/**\n * Generate a summary of abandoned branch entries.\n *\n * @param entries - Session entries to summarize (chronological order)\n * @param options - Generation options\n */\nasync function completeBranchSummary(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n): Promise<AssistantMessage> {\n\tif (!streamFn) return completeSimple(model, context, options);\n\tconst stream = await streamFn(model, context, options);\n\treturn stream.result();\n}\n\nexport async function generateBranchSummary(\n\tentries: SessionEntry[],\n\toptions: GenerateBranchSummaryOptions,\n): Promise<BranchSummaryResult> {\n\tconst {\n\t\tmodel,\n\t\tapiKey,\n\t\theaders,\n\t\tsignal,\n\t\tcustomInstructions,\n\t\treplaceInstructions,\n\t\treserveTokens = 16384,\n\t\tstreamFn,\n\t} = options;\n\n\t// Token budget = context window minus reserved space for prompt + response\n\tconst contextWindow = model.contextWindow || 128000;\n\tconst tokenBudget = contextWindow - reserveTokens;\n\n\tconst { messages, fileOps } = prepareBranchEntries(entries, tokenBudget);\n\n\tif (messages.length === 0) {\n\t\treturn { summary: \"No content to summarize\" };\n\t}\n\n\t// Transform to LLM-compatible messages, then serialize to text\n\t// Serialization prevents the model from treating it as a conversation to continue\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build prompt\n\tlet instructions: string;\n\tif (replaceInstructions && customInstructions) {\n\t\tinstructions = customInstructions;\n\t} else if (customInstructions) {\n\t\tinstructions = `${BRANCH_SUMMARY_PROMPT}\\n\\nAdditional focus: ${customInstructions}`;\n\t} else {\n\t\tinstructions = BRANCH_SUMMARY_PROMPT;\n\t}\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${instructions}`;\n\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\t// Call LLM for summarization. Retry transient provider/watchdog failures so branch summaries\n\t// use the same reliability doctrine as compaction instead of bypassing the host stream wrapper.\n\tlet response: AssistantMessage | undefined;\n\tfor (let attempt = 0; attempt < 3; attempt++) {\n\t\tresponse = await completeBranchSummary(\n\t\t\tmodel,\n\t\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\t\t{ apiKey, headers, signal, maxTokens: 2048 },\n\t\t\tstreamFn,\n\t\t);\n\t\tif (response.stopReason !== \"error\") break;\n\t\tconst classified = classifyFailure({ message: response.errorMessage ?? \"\", provider: response.provider });\n\t\tif (!classified.retryable || signal.aborted) break;\n\t}\n\tif (!response) return { error: \"Summarization failed\" };\n\n\t// Check if aborted or errored\n\tif (response.stopReason === \"aborted\") {\n\t\treturn { aborted: true };\n\t}\n\tif (response.stopReason === \"error\") {\n\t\treturn { error: response.errorMessage || \"Summarization failed\" };\n\t}\n\n\tlet summary = response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n\n\t// Prepend preamble to provide context about the branch summary\n\tsummary = BRANCH_SUMMARY_PREAMBLE + summary;\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\treturn {\n\t\tsummary: summary || \"No summary generated\",\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}

package/dist/compaction/branch-summarization.js

@@ -6,6 +6,7 @@
  */
 import { completeSimple } from "@caupulican/pi-ai";
 import { convertToLlm, createBranchSummaryMessage, createCompactionSummaryMessage, createCustomMessage, } from "../messages.js";
+import { classifyFailure } from "../reliability/classifier.js";
 import { estimateTokens } from "./compaction.js";
 import { computeFileLists, createFileOps, extractFileOpsFromMessage, formatFileOperations, SUMMARIZATION_SYSTEM_PROMPT, serializeConversation, } from "./utils.js";
 // ============================================================================
@@ -184,8 +185,14 @@ Keep each section concise. Preserve exact file paths, function names, and error
  * @param entries - Session entries to summarize (chronological order)
  * @param options - Generation options
  */
+async function completeBranchSummary(model, context, options, streamFn) {
+    if (!streamFn)
+        return completeSimple(model, context, options);
+    const stream = await streamFn(model, context, options);
+    return stream.result();
+}
 export async function generateBranchSummary(entries, options) {
-    const { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;
+    const { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384, streamFn, } = options;
     // Token budget = context window minus reserved space for prompt + response
     const contextWindow = model.contextWindow || 128000;
     const tokenBudget = contextWindow - reserveTokens;
@@ -216,8 +223,19 @@ export async function generateBranchSummary(entries, options) {
             timestamp: Date.now(),
         },
     ];
-    // Call LLM for summarization
-    const response = await completeSimple(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, { apiKey, headers, signal, maxTokens: 2048 });
+    // Call LLM for summarization. Retry transient provider/watchdog failures so branch summaries
+    // use the same reliability doctrine as compaction instead of bypassing the host stream wrapper.
+    let response;
+    for (let attempt = 0; attempt < 3; attempt++) {
+        response = await completeBranchSummary(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, { apiKey, headers, signal, maxTokens: 2048 }, streamFn);
+        if (response.stopReason !== "error")
+            break;
+        const classified = classifyFailure({ message: response.errorMessage ?? "", provider: response.provider });
+        if (!classified.retryable || signal.aborted)
+            break;
+    }
+    if (!response)
+        return { error: "Summarization failed" };
     // Check if aborted or errored
     if (response.stopReason === "aborted") {
         return { aborted: true };

package/dist/compaction/branch-summarization.js.map

@@ -1 +1 @@
-{"version":3,"file":"branch-summarization.js","sourceRoot":"","sources":["../../src/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,EACN,YAAY,EACZ,0BAA0B,EAC1B,8BAA8B,EAC9B,mBAAmB,GACnB,MAAM,gBAAgB,CAAC;AAGxB,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,EACN,gBAAgB,EAChB,aAAa,EACb,yBAAyB,EAEzB,oBAAoB,EACpB,2BAA2B,EAC3B,qBAAqB,GACrB,MAAM,YAAY,CAAC;AAuDpB,+EAA+E;AAC/E,mBAAmB;AACnB,+EAA+E;AAE/E;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAA+B,EAC/B,SAAwB,EACxB,QAAgB,EACO;IACvB,2CAA2C;IAC3C,IAAI,CAAC,SAAS,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC;IAChD,CAAC;IAED,2DAA2D;IAC3D,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvE,MAAM,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;IAE/C,iFAAiF;IACjF,IAAI,gBAAgB,GAAkB,IAAI,CAAC;IAC3C,KAAK,IAAI,CAAC,GAAG,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACjD,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;YACnC,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACpC,MAAM;QACP,CAAC;IACF,CAAC;IAED,wDAAwD;IACxD,MAAM,OAAO,GAAmB,EAAE,CAAC;IACnC,IAAI,OAAO,GAAkB,SAAS,CAAC;IAEvC,OAAO,OAAO,IAAI,OAAO,KAAK,gBAAgB,EAAE,CAAC;QAChD,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK;YAAE,MAAM;QAClB,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACpB,OAAO,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC1B,CAAC;IAED,qCAAqC;IACrC,OAAO,CAAC,OAAO,EAAE,CAAC;IAElB,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,CAAC;AAAA,CACrC;AAED,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E;;;GAGG;AACH,SAAS,mBAAmB,CAAC,KAAmB,EAA4B;IAC3E,QAAQ,KAAK,CAAC,IAAI,EAAE,CAAC;QACpB,KAAK,SAAS;YACb,0DAA0D;YAC1D,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,KAAK,YAAY;gBAAE,OAAO,SAAS,CAAC;YAC1D,OAAO,KAAK,CAAC,OAAO,CAAC;QAEtB,KAAK,gBAAgB;YACpB,OAAO,mBAAmB,CAAC,KAAK,CAAC,UAAU,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAE5G,KAAK,gBAAgB;YACpB,OAAO,0BAA0B,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAEjF,KAAK,YAAY;YAChB,OAAO,8BAA8B,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,YAAY,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAE3F,iDAAiD;QACjD,KAAK,uBAAuB,CAAC;QAC7B,KAAK,cAAc,CAAC;QACpB,KAAK,QAAQ,CAAC;QACd,KAAK,OAAO,CAAC;QACb,KAAK,cAAc;YAClB,OAAO,SAAS,CAAC;IACnB,CAAC;AAAA,CACD;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,oBAAoB,CAAC,OAAuB,EAAE,WAAW,GAAW,CAAC,EAAqB;IACzG,MAAM,QAAQ,GAAmB,EAAE,CAAC;IACpC,MAAM,OAAO,GAAG,aAAa,EAAE,CAAC;IAChC,IAAI,WAAW,GAAG,CAAC,CAAC;IAEpB,yFAAyF;IACzF,gFAAgF;IAChF,6FAA6F;IAC7F,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YACzE,MAAM,OAAO,GAAG,KAAK,CAAC,OAA+B,CAAC;YACtD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;gBACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,SAAS;oBAAE,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;YACxD,CAAC;YACD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,EAAE,CAAC;gBAC1C,0EAA0E;gBAC1E,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,aAAa,EAAE,CAAC;oBACvC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;gBACvB,CAAC;YACF,CAAC;QACF,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,KAAK,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,MAAM,OAAO,GAAG,mBAAmB,CAAC,KAAK,CAAC,CAAC;QAC3C,IAAI,CAAC,OAAO;YAAE,SAAS;QAEvB,wDAAwD;QACxD,yBAAyB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAE5C,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;QAEvC,6BAA6B;QAC7B,IAAI,WAAW,GAAG,CAAC,IAAI,WAAW,GAAG,MAAM,GAAG,WAAW,EAAE,CAAC;YAC3D,6EAA6E;YAC7E,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;gBACpE,IAAI,WAAW,GAAG,WAAW,GAAG,GAAG,EAAE,CAAC;oBACrC,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;oBAC1B,WAAW,IAAI,MAAM,CAAC;gBACvB,CAAC;YACF,CAAC;YACD,8BAA8B;YAC9B,MAAM;QACP,CAAC;QAED,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAC1B,WAAW,IAAI,MAAM,CAAC;IACvB,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC;AAAA,CAC1C;AAED,+EAA+E;AAC/E,qBAAqB;AACrB,+EAA+E;AAE/E,MAAM,uBAAuB,GAAG;;;CAG/B,CAAC;AAEF,MAAM,qBAAqB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;0FA2B4D,CAAC;AAE3F;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAC1C,OAAuB,EACvB,OAAqC,EACN;IAC/B,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,kBAAkB,EAAE,mBAAmB,EAAE,aAAa,GAAG,KAAK,EAAE,GAAG,OAAO,CAAC;IAEnH,2EAA2E;IAC3E,MAAM,aAAa,GAAG,KAAK,CAAC,aAAa,IAAI,MAAM,CAAC;IACpD,MAAM,WAAW,GAAG,aAAa,GAAG,aAAa,CAAC;IAElD,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,oBAAoB,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAEzE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,OAAO,EAAE,OAAO,EAAE,yBAAyB,EAAE,CAAC;IAC/C,CAAC;IAED,+DAA+D;IAC/D,kFAAkF;IAClF,MAAM,WAAW,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC3C,MAAM,gBAAgB,GAAG,qBAAqB,CAAC,WAAW,CAAC,CAAC;IAE5D,eAAe;IACf,IAAI,YAAoB,CAAC;IACzB,IAAI,mBAAmB,IAAI,kBAAkB,EAAE,CAAC;QAC/C,YAAY,GAAG,kBAAkB,CAAC;IACnC,CAAC;SAAM,IAAI,kBAAkB,EAAE,CAAC;QAC/B,YAAY,GAAG,GAAG,qBAAqB,yBAAyB,kBAAkB,EAAE,CAAC;IACtF,CAAC;SAAM,CAAC;QACP,YAAY,GAAG,qBAAqB,CAAC;IACtC,CAAC;IACD,MAAM,UAAU,GAAG,mBAAmB,gBAAgB,wBAAwB,YAAY,EAAE,CAAC;IAE7F,MAAM,qBAAqB,GAAG;QAC7B;YACC,IAAI,EAAE,MAAe;YACrB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;YACtD,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB;KACD,CAAC;IAEF,6BAA6B;IAC7B,MAAM,QAAQ,GAAG,MAAM,cAAc,CACpC,KAAK,EACL,EAAE,YAAY,EAAE,2BAA2B,EAAE,QAAQ,EAAE,qBAAqB,EAAE,EAC9E,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,CAC5C,CAAC;IAEF,8BAA8B;IAC9B,IAAI,QAAQ,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC1B,CAAC;IACD,IAAI,QAAQ,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;QACrC,OAAO,EAAE,KAAK,EAAE,QAAQ,CAAC,YAAY,IAAI,sBAAsB,EAAE,CAAC;IACnE,CAAC;IAED,IAAI,OAAO,GAAG,QAAQ,CAAC,OAAO;SAC5B,MAAM,CAAC,CAAC,CAAC,EAAuC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC;SACrE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;SAClB,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,+DAA+D;IAC/D,OAAO,GAAG,uBAAuB,GAAG,OAAO,CAAC;IAE5C,2CAA2C;IAC3C,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC/D,OAAO,IAAI,oBAAoB,CAAC,SAAS,EAAE,aAAa,CAAC,CAAC;IAE1D,OAAO;QACN,OAAO,EAAE,OAAO,IAAI,sBAAsB;QAC1C,SAAS;QACT,aAAa;KACb,CAAC;AAAA,CACF","sourcesContent":["/**\n * Branch summarization for tree navigation.\n *\n * When navigating to a different point in the session tree, this generates\n * a summary of the branch being left so context isn't lost.\n */\n\nimport type { Model } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage } from \"../types.ts\";\nimport { estimateTokens } from \"./compaction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface BranchSummaryResult {\n\tsummary?: string;\n\treadFiles?: string[];\n\tmodifiedFiles?: string[];\n\taborted?: boolean;\n\terror?: string;\n}\n\n/** Details stored in BranchSummaryEntry.details for file tracking */\nexport interface BranchSummaryDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\nexport type { FileOperations } from \"./utils.ts\";\n\nexport interface BranchPreparation {\n\t/** Messages extracted for summarization, in chronological order */\n\tmessages: AgentMessage[];\n\t/** File operations extracted from tool calls */\n\tfileOps: FileOperations;\n\t/** Total estimated tokens in messages */\n\ttotalTokens: number;\n}\n\nexport interface CollectEntriesResult {\n\t/** Entries to summarize, in chronological order */\n\tentries: SessionEntry[];\n\t/** Common ancestor between old and new position, if any */\n\tcommonAncestorId: string | null;\n}\n\nexport interface GenerateBranchSummaryOptions {\n\t/** Model to use for summarization */\n\tmodel: Model<any>;\n\t/** API key for the model */\n\tapiKey: string;\n\t/** Request headers for the model */\n\theaders?: Record<string, string>;\n\t/** Abort signal for cancellation */\n\tsignal: AbortSignal;\n\t/** Optional custom instructions for summarization */\n\tcustomInstructions?: string;\n\t/** If true, customInstructions replaces the default prompt instead of being appended */\n\treplaceInstructions?: boolean;\n\t/** Tokens reserved for prompt + LLM response (default 16384) */\n\treserveTokens?: number;\n}\n\n// ============================================================================\n// Entry Collection\n// ============================================================================\n\n/**\n * Collect entries that should be summarized when navigating from one position to another.\n *\n * Walks from oldLeafId back to the common ancestor with targetId, collecting entries\n * along the way. Does NOT stop at compaction boundaries - those are included and their\n * summaries become context.\n *\n * @param session - Session manager (read-only access)\n * @param oldLeafId - Current position (where we're navigating from)\n * @param targetId - Target position (where we're navigating to)\n * @returns Entries to summarize and the common ancestor\n */\nexport function collectEntriesForBranchSummary(\n\tsession: ReadonlySessionManager,\n\toldLeafId: string | null,\n\ttargetId: string,\n): CollectEntriesResult {\n\t// If no old position, nothing to summarize\n\tif (!oldLeafId) {\n\t\treturn { entries: [], commonAncestorId: null };\n\t}\n\n\t// Find common ancestor (deepest node that's on both paths)\n\tconst oldPath = new Set(session.getBranch(oldLeafId).map((e) => e.id));\n\tconst targetPath = session.getBranch(targetId);\n\n\t// targetPath is root-first, so iterate backwards to find deepest common ancestor\n\tlet commonAncestorId: string | null = null;\n\tfor (let i = targetPath.length - 1; i >= 0; i--) {\n\t\tif (oldPath.has(targetPath[i].id)) {\n\t\t\tcommonAncestorId = targetPath[i].id;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Collect entries from old leaf back to common ancestor\n\tconst entries: SessionEntry[] = [];\n\tlet current: string | null = oldLeafId;\n\n\twhile (current && current !== commonAncestorId) {\n\t\tconst entry = session.getEntry(current);\n\t\tif (!entry) break;\n\t\tentries.push(entry);\n\t\tcurrent = entry.parentId;\n\t}\n\n\t// Reverse to get chronological order\n\tentries.reverse();\n\n\treturn { entries, commonAncestorId };\n}\n\n// ============================================================================\n// Entry to Message Conversion\n// ============================================================================\n\n/**\n * Extract AgentMessage from a session entry.\n * Similar to getMessageFromEntry in compaction.ts but also handles compaction entries.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tswitch (entry.type) {\n\t\tcase \"message\":\n\t\t\t// Skip tool results - context is in assistant's tool call\n\t\t\tif (entry.message.role === \"toolResult\") return undefined;\n\t\t\treturn entry.message;\n\n\t\tcase \"custom_message\":\n\t\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\n\t\tcase \"branch_summary\":\n\t\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\n\t\tcase \"compaction\":\n\t\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\n\t\t// These don't contribute to conversation content\n\t\tcase \"thinking_level_change\":\n\t\tcase \"model_change\":\n\t\tcase \"custom\":\n\t\tcase \"label\":\n\t\tcase \"session_info\":\n\t\t\treturn undefined;\n\t}\n}\n\n/**\n * Prepare entries for summarization with token budget.\n *\n * Walks entries from NEWEST to OLDEST, adding messages until we hit the token budget.\n * This ensures we keep the most recent context when the branch is too long.\n *\n * Also collects file operations from:\n * - Tool calls in assistant messages\n * - Existing branch_summary entries' details (for cumulative tracking)\n *\n * @param entries - Entries in chronological order\n * @param tokenBudget - Maximum tokens to include (0 = no limit)\n */\nexport function prepareBranchEntries(entries: SessionEntry[], tokenBudget: number = 0): BranchPreparation {\n\tconst messages: AgentMessage[] = [];\n\tconst fileOps = createFileOps();\n\tlet totalTokens = 0;\n\n\t// First pass: collect file ops from ALL entries (even if they don't fit in token budget)\n\t// This ensures we capture cumulative file tracking from nested branch summaries\n\t// Only extract from pi-generated summaries (fromHook !== true), not extension-generated ones\n\tfor (const entry of entries) {\n\t\tif (entry.type === \"branch_summary\" && !entry.fromHook && entry.details) {\n\t\t\tconst details = entry.details as BranchSummaryDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\t// Modified files go into both edited and written for proper deduplication\n\t\t\t\tfor (const f of details.modifiedFiles) {\n\t\t\t\t\tfileOps.edited.add(f);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Second pass: walk from newest to oldest, adding messages until token budget\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tconst message = getMessageFromEntry(entry);\n\t\tif (!message) continue;\n\n\t\t// Extract file ops from assistant messages (tool calls)\n\t\textractFileOpsFromMessage(message, fileOps);\n\n\t\tconst tokens = estimateTokens(message);\n\n\t\t// Check budget before adding\n\t\tif (tokenBudget > 0 && totalTokens + tokens > tokenBudget) {\n\t\t\t// If this is a summary entry, try to fit it anyway as it's important context\n\t\t\tif (entry.type === \"compaction\" || entry.type === \"branch_summary\") {\n\t\t\t\tif (totalTokens < tokenBudget * 0.9) {\n\t\t\t\t\tmessages.unshift(message);\n\t\t\t\t\ttotalTokens += tokens;\n\t\t\t\t}\n\t\t\t}\n\t\t\t// Stop - we've hit the budget\n\t\t\tbreak;\n\t\t}\n\n\t\tmessages.unshift(message);\n\t\ttotalTokens += tokens;\n\t}\n\n\treturn { messages, fileOps, totalTokens };\n}\n\n// ============================================================================\n// Summary Generation\n// ============================================================================\n\nconst BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.\nSummary of that exploration:\n\n`;\n\nconst BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.\n\nUse this EXACT format:\n\n## Goal\n[What was the user trying to accomplish in this branch?]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Work that was started but not finished]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [What should happen next to continue this work]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\n/**\n * Generate a summary of abandoned branch entries.\n *\n * @param entries - Session entries to summarize (chronological order)\n * @param options - Generation options\n */\nexport async function generateBranchSummary(\n\tentries: SessionEntry[],\n\toptions: GenerateBranchSummaryOptions,\n): Promise<BranchSummaryResult> {\n\tconst { model, apiKey, headers, signal, customInstructions, replaceInstructions, reserveTokens = 16384 } = options;\n\n\t// Token budget = context window minus reserved space for prompt + response\n\tconst contextWindow = model.contextWindow || 128000;\n\tconst tokenBudget = contextWindow - reserveTokens;\n\n\tconst { messages, fileOps } = prepareBranchEntries(entries, tokenBudget);\n\n\tif (messages.length === 0) {\n\t\treturn { summary: \"No content to summarize\" };\n\t}\n\n\t// Transform to LLM-compatible messages, then serialize to text\n\t// Serialization prevents the model from treating it as a conversation to continue\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build prompt\n\tlet instructions: string;\n\tif (replaceInstructions && customInstructions) {\n\t\tinstructions = customInstructions;\n\t} else if (customInstructions) {\n\t\tinstructions = `${BRANCH_SUMMARY_PROMPT}\\n\\nAdditional focus: ${customInstructions}`;\n\t} else {\n\t\tinstructions = BRANCH_SUMMARY_PROMPT;\n\t}\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${instructions}`;\n\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\t// Call LLM for summarization\n\tconst response = await completeSimple(\n\t\tmodel,\n\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\t{ apiKey, headers, signal, maxTokens: 2048 },\n\t);\n\n\t// Check if aborted or errored\n\tif (response.stopReason === \"aborted\") {\n\t\treturn { aborted: true };\n\t}\n\tif (response.stopReason === \"error\") {\n\t\treturn { error: response.errorMessage || \"Summarization failed\" };\n\t}\n\n\tlet summary = response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n\n\t// Prepend preamble to provide context about the branch summary\n\tsummary = BRANCH_SUMMARY_PREAMBLE + summary;\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\treturn {\n\t\tsummary: summary || \"No summary generated\",\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}
+{"version":3,"file":"branch-summarization.js","sourceRoot":"","sources":["../../src/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,OAAO,EACN,YAAY,EACZ,0BAA0B,EAC1B,8BAA8B,EAC9B,mBAAmB,GACnB,MAAM,gBAAgB,CAAC;AACxB,OAAO,EAAE,eAAe,EAAE,MAAM,8BAA8B,CAAC;AAG/D,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,EACN,gBAAgB,EAChB,aAAa,EACb,yBAAyB,EAEzB,oBAAoB,EACpB,2BAA2B,EAC3B,qBAAqB,GACrB,MAAM,YAAY,CAAC;AAyDpB,+EAA+E;AAC/E,mBAAmB;AACnB,+EAA+E;AAE/E;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,8BAA8B,CAC7C,OAA+B,EAC/B,SAAwB,EACxB,QAAgB,EACO;IACvB,2CAA2C;IAC3C,IAAI,CAAC,SAAS,EAAE,CAAC;QAChB,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC;IAChD,CAAC;IAED,2DAA2D;IAC3D,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvE,MAAM,UAAU,GAAG,OAAO,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;IAE/C,iFAAiF;IACjF,IAAI,gBAAgB,GAAkB,IAAI,CAAC;IAC3C,KAAK,IAAI,CAAC,GAAG,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QACjD,IAAI,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC;YACnC,gBAAgB,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACpC,MAAM;QACP,CAAC;IACF,CAAC;IAED,wDAAwD;IACxD,MAAM,OAAO,GAAmB,EAAE,CAAC;IACnC,IAAI,OAAO,GAAkB,SAAS,CAAC;IAEvC,OAAO,OAAO,IAAI,OAAO,KAAK,gBAAgB,EAAE,CAAC;QAChD,MAAM,KAAK,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK;YAAE,MAAM;QAClB,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACpB,OAAO,GAAG,KAAK,CAAC,QAAQ,CAAC;IAC1B,CAAC;IAED,qCAAqC;IACrC,OAAO,CAAC,OAAO,EAAE,CAAC;IAElB,OAAO,EAAE,OAAO,EAAE,gBAAgB,EAAE,CAAC;AAAA,CACrC;AAED,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E;;;GAGG;AACH,SAAS,mBAAmB,CAAC,KAAmB,EAA4B;IAC3E,QAAQ,KAAK,CAAC,IAAI,EAAE,CAAC;QACpB,KAAK,SAAS;YACb,0DAA0D;YAC1D,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,KAAK,YAAY;gBAAE,OAAO,SAAS,CAAC;YAC1D,OAAO,KAAK,CAAC,OAAO,CAAC;QAEtB,KAAK,gBAAgB;YACpB,OAAO,mBAAmB,CAAC,KAAK,CAAC,UAAU,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAE5G,KAAK,gBAAgB;YACpB,OAAO,0BAA0B,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAEjF,KAAK,YAAY;YAChB,OAAO,8BAA8B,CAAC,KAAK,CAAC,OAAO,EAAE,KAAK,CAAC,YAAY,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QAE3F,iDAAiD;QACjD,KAAK,uBAAuB,CAAC;QAC7B,KAAK,cAAc,CAAC;QACpB,KAAK,QAAQ,CAAC;QACd,KAAK,OAAO,CAAC;QACb,KAAK,cAAc;YAClB,OAAO,SAAS,CAAC;IACnB,CAAC;AAAA,CACD;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,oBAAoB,CAAC,OAAuB,EAAE,WAAW,GAAW,CAAC,EAAqB;IACzG,MAAM,QAAQ,GAAmB,EAAE,CAAC;IACpC,MAAM,OAAO,GAAG,aAAa,EAAE,CAAC;IAChC,IAAI,WAAW,GAAG,CAAC,CAAC;IAEpB,yFAAyF;IACzF,gFAAgF;IAChF,6FAA6F;IAC7F,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,KAAK,CAAC,OAAO,EAAE,CAAC;YACzE,MAAM,OAAO,GAAG,KAAK,CAAC,OAA+B,CAAC;YACtD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;gBACtC,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,SAAS;oBAAE,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;YACxD,CAAC;YACD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,CAAC,EAAE,CAAC;gBAC1C,0EAA0E;gBAC1E,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,aAAa,EAAE,CAAC;oBACvC,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;gBACvB,CAAC;YACF,CAAC;QACF,CAAC;IACF,CAAC;IAED,8EAA8E;IAC9E,KAAK,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,MAAM,OAAO,GAAG,mBAAmB,CAAC,KAAK,CAAC,CAAC;QAC3C,IAAI,CAAC,OAAO;YAAE,SAAS;QAEvB,wDAAwD;QACxD,yBAAyB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAE5C,MAAM,MAAM,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;QAEvC,6BAA6B;QAC7B,IAAI,WAAW,GAAG,CAAC,IAAI,WAAW,GAAG,MAAM,GAAG,WAAW,EAAE,CAAC;YAC3D,6EAA6E;YAC7E,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;gBACpE,IAAI,WAAW,GAAG,WAAW,GAAG,GAAG,EAAE,CAAC;oBACrC,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;oBAC1B,WAAW,IAAI,MAAM,CAAC;gBACvB,CAAC;YACF,CAAC;YACD,8BAA8B;YAC9B,MAAM;QACP,CAAC;QAED,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAC1B,WAAW,IAAI,MAAM,CAAC;IACvB,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC;AAAA,CAC1C;AAED,+EAA+E;AAC/E,qBAAqB;AACrB,+EAA+E;AAE/E,MAAM,uBAAuB,GAAG;;;CAG/B,CAAC;AAEF,MAAM,qBAAqB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;0FA2B4D,CAAC;AAE3F;;;;;GAKG;AACH,KAAK,UAAU,qBAAqB,CACnC,KAAiB,EACjB,OAAgB,EAChB,OAA4B,EAC5B,QAAmB,EACS;IAC5B,IAAI,CAAC,QAAQ;QAAE,OAAO,cAAc,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IAC9D,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC;IACvD,OAAO,MAAM,CAAC,MAAM,EAAE,CAAC;AAAA,CACvB;AAED,MAAM,CAAC,KAAK,UAAU,qBAAqB,CAC1C,OAAuB,EACvB,OAAqC,EACN;IAC/B,MAAM,EACL,KAAK,EACL,MAAM,EACN,OAAO,EACP,MAAM,EACN,kBAAkB,EAClB,mBAAmB,EACnB,aAAa,GAAG,KAAK,EACrB,QAAQ,GACR,GAAG,OAAO,CAAC;IAEZ,2EAA2E;IAC3E,MAAM,aAAa,GAAG,KAAK,CAAC,aAAa,IAAI,MAAM,CAAC;IACpD,MAAM,WAAW,GAAG,aAAa,GAAG,aAAa,CAAC;IAElD,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,GAAG,oBAAoB,CAAC,OAAO,EAAE,WAAW,CAAC,CAAC;IAEzE,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3B,OAAO,EAAE,OAAO,EAAE,yBAAyB,EAAE,CAAC;IAC/C,CAAC;IAED,+DAA+D;IAC/D,kFAAkF;IAClF,MAAM,WAAW,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC3C,MAAM,gBAAgB,GAAG,qBAAqB,CAAC,WAAW,CAAC,CAAC;IAE5D,eAAe;IACf,IAAI,YAAoB,CAAC;IACzB,IAAI,mBAAmB,IAAI,kBAAkB,EAAE,CAAC;QAC/C,YAAY,GAAG,kBAAkB,CAAC;IACnC,CAAC;SAAM,IAAI,kBAAkB,EAAE,CAAC;QAC/B,YAAY,GAAG,GAAG,qBAAqB,yBAAyB,kBAAkB,EAAE,CAAC;IACtF,CAAC;SAAM,CAAC;QACP,YAAY,GAAG,qBAAqB,CAAC;IACtC,CAAC;IACD,MAAM,UAAU,GAAG,mBAAmB,gBAAgB,wBAAwB,YAAY,EAAE,CAAC;IAE7F,MAAM,qBAAqB,GAAG;QAC7B;YACC,IAAI,EAAE,MAAe;YACrB,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAe,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;YACtD,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;SACrB;KACD,CAAC;IAEF,6FAA6F;IAC7F,gGAAgG;IAChG,IAAI,QAAsC,CAAC;IAC3C,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,EAAE,EAAE,CAAC;QAC9C,QAAQ,GAAG,MAAM,qBAAqB,CACrC,KAAK,EACL,EAAE,YAAY,EAAE,2BAA2B,EAAE,QAAQ,EAAE,qBAAqB,EAAE,EAC9E,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,EAC5C,QAAQ,CACR,CAAC;QACF,IAAI,QAAQ,CAAC,UAAU,KAAK,OAAO;YAAE,MAAM;QAC3C,MAAM,UAAU,GAAG,eAAe,CAAC,EAAE,OAAO,EAAE,QAAQ,CAAC,YAAY,IAAI,EAAE,EAAE,QAAQ,EAAE,QAAQ,CAAC,QAAQ,EAAE,CAAC,CAAC;QAC1G,IAAI,CAAC,UAAU,CAAC,SAAS,IAAI,MAAM,CAAC,OAAO;YAAE,MAAM;IACpD,CAAC;IACD,IAAI,CAAC,QAAQ;QAAE,OAAO,EAAE,KAAK,EAAE,sBAAsB,EAAE,CAAC;IAExD,8BAA8B;IAC9B,IAAI,QAAQ,CAAC,UAAU,KAAK,SAAS,EAAE,CAAC;QACvC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAC1B,CAAC;IACD,IAAI,QAAQ,CAAC,UAAU,KAAK,OAAO,EAAE,CAAC;QACrC,OAAO,EAAE,KAAK,EAAE,QAAQ,CAAC,YAAY,IAAI,sBAAsB,EAAE,CAAC;IACnE,CAAC;IAED,IAAI,OAAO,GAAG,QAAQ,CAAC,OAAO;SAC5B,MAAM,CAAC,CAAC,CAAC,EAAuC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC;SACrE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;SAClB,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,+DAA+D;IAC/D,OAAO,GAAG,uBAAuB,GAAG,OAAO,CAAC;IAE5C,2CAA2C;IAC3C,MAAM,EAAE,SAAS,EAAE,aAAa,EAAE,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;IAC/D,OAAO,IAAI,oBAAoB,CAAC,SAAS,EAAE,aAAa,CAAC,CAAC;IAE1D,OAAO;QACN,OAAO,EAAE,OAAO,IAAI,sBAAsB;QAC1C,SAAS;QACT,aAAa;KACb,CAAC;AAAA,CACF","sourcesContent":["/**\n * Branch summarization for tree navigation.\n *\n * When navigating to a different point in the session tree, this generates\n * a summary of the branch being left so context isn't lost.\n */\n\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport { classifyFailure } from \"../reliability/classifier.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage, StreamFn } from \"../types.ts\";\nimport { estimateTokens } from \"./compaction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface BranchSummaryResult {\n\tsummary?: string;\n\treadFiles?: string[];\n\tmodifiedFiles?: string[];\n\taborted?: boolean;\n\terror?: string;\n}\n\n/** Details stored in BranchSummaryEntry.details for file tracking */\nexport interface BranchSummaryDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\nexport type { FileOperations } from \"./utils.ts\";\n\nexport interface BranchPreparation {\n\t/** Messages extracted for summarization, in chronological order */\n\tmessages: AgentMessage[];\n\t/** File operations extracted from tool calls */\n\tfileOps: FileOperations;\n\t/** Total estimated tokens in messages */\n\ttotalTokens: number;\n}\n\nexport interface CollectEntriesResult {\n\t/** Entries to summarize, in chronological order */\n\tentries: SessionEntry[];\n\t/** Common ancestor between old and new position, if any */\n\tcommonAncestorId: string | null;\n}\n\nexport interface GenerateBranchSummaryOptions {\n\t/** Model to use for summarization */\n\tmodel: Model<any>;\n\t/** API key for the model */\n\tapiKey: string;\n\t/** Request headers for the model */\n\theaders?: Record<string, string>;\n\t/** Abort signal for cancellation */\n\tsignal: AbortSignal;\n\t/** Optional custom instructions for summarization */\n\tcustomInstructions?: string;\n\t/** If true, customInstructions replaces the default prompt instead of being appended */\n\treplaceInstructions?: boolean;\n\t/** Tokens reserved for prompt + LLM response (default 16384) */\n\treserveTokens?: number;\n\t/** Optional wrapped stream function (watchdog/retry-capable host path). */\n\tstreamFn?: StreamFn;\n}\n\n// ============================================================================\n// Entry Collection\n// ============================================================================\n\n/**\n * Collect entries that should be summarized when navigating from one position to another.\n *\n * Walks from oldLeafId back to the common ancestor with targetId, collecting entries\n * along the way. Does NOT stop at compaction boundaries - those are included and their\n * summaries become context.\n *\n * @param session - Session manager (read-only access)\n * @param oldLeafId - Current position (where we're navigating from)\n * @param targetId - Target position (where we're navigating to)\n * @returns Entries to summarize and the common ancestor\n */\nexport function collectEntriesForBranchSummary(\n\tsession: ReadonlySessionManager,\n\toldLeafId: string | null,\n\ttargetId: string,\n): CollectEntriesResult {\n\t// If no old position, nothing to summarize\n\tif (!oldLeafId) {\n\t\treturn { entries: [], commonAncestorId: null };\n\t}\n\n\t// Find common ancestor (deepest node that's on both paths)\n\tconst oldPath = new Set(session.getBranch(oldLeafId).map((e) => e.id));\n\tconst targetPath = session.getBranch(targetId);\n\n\t// targetPath is root-first, so iterate backwards to find deepest common ancestor\n\tlet commonAncestorId: string | null = null;\n\tfor (let i = targetPath.length - 1; i >= 0; i--) {\n\t\tif (oldPath.has(targetPath[i].id)) {\n\t\t\tcommonAncestorId = targetPath[i].id;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Collect entries from old leaf back to common ancestor\n\tconst entries: SessionEntry[] = [];\n\tlet current: string | null = oldLeafId;\n\n\twhile (current && current !== commonAncestorId) {\n\t\tconst entry = session.getEntry(current);\n\t\tif (!entry) break;\n\t\tentries.push(entry);\n\t\tcurrent = entry.parentId;\n\t}\n\n\t// Reverse to get chronological order\n\tentries.reverse();\n\n\treturn { entries, commonAncestorId };\n}\n\n// ============================================================================\n// Entry to Message Conversion\n// ============================================================================\n\n/**\n * Extract AgentMessage from a session entry.\n * Similar to getMessageFromEntry in compaction.ts but also handles compaction entries.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tswitch (entry.type) {\n\t\tcase \"message\":\n\t\t\t// Skip tool results - context is in assistant's tool call\n\t\t\tif (entry.message.role === \"toolResult\") return undefined;\n\t\t\treturn entry.message;\n\n\t\tcase \"custom_message\":\n\t\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\n\t\tcase \"branch_summary\":\n\t\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\n\t\tcase \"compaction\":\n\t\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\n\t\t// These don't contribute to conversation content\n\t\tcase \"thinking_level_change\":\n\t\tcase \"model_change\":\n\t\tcase \"custom\":\n\t\tcase \"label\":\n\t\tcase \"session_info\":\n\t\t\treturn undefined;\n\t}\n}\n\n/**\n * Prepare entries for summarization with token budget.\n *\n * Walks entries from NEWEST to OLDEST, adding messages until we hit the token budget.\n * This ensures we keep the most recent context when the branch is too long.\n *\n * Also collects file operations from:\n * - Tool calls in assistant messages\n * - Existing branch_summary entries' details (for cumulative tracking)\n *\n * @param entries - Entries in chronological order\n * @param tokenBudget - Maximum tokens to include (0 = no limit)\n */\nexport function prepareBranchEntries(entries: SessionEntry[], tokenBudget: number = 0): BranchPreparation {\n\tconst messages: AgentMessage[] = [];\n\tconst fileOps = createFileOps();\n\tlet totalTokens = 0;\n\n\t// First pass: collect file ops from ALL entries (even if they don't fit in token budget)\n\t// This ensures we capture cumulative file tracking from nested branch summaries\n\t// Only extract from pi-generated summaries (fromHook !== true), not extension-generated ones\n\tfor (const entry of entries) {\n\t\tif (entry.type === \"branch_summary\" && !entry.fromHook && entry.details) {\n\t\t\tconst details = entry.details as BranchSummaryDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\t// Modified files go into both edited and written for proper deduplication\n\t\t\t\tfor (const f of details.modifiedFiles) {\n\t\t\t\t\tfileOps.edited.add(f);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t}\n\n\t// Second pass: walk from newest to oldest, adding messages until token budget\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tconst message = getMessageFromEntry(entry);\n\t\tif (!message) continue;\n\n\t\t// Extract file ops from assistant messages (tool calls)\n\t\textractFileOpsFromMessage(message, fileOps);\n\n\t\tconst tokens = estimateTokens(message);\n\n\t\t// Check budget before adding\n\t\tif (tokenBudget > 0 && totalTokens + tokens > tokenBudget) {\n\t\t\t// If this is a summary entry, try to fit it anyway as it's important context\n\t\t\tif (entry.type === \"compaction\" || entry.type === \"branch_summary\") {\n\t\t\t\tif (totalTokens < tokenBudget * 0.9) {\n\t\t\t\t\tmessages.unshift(message);\n\t\t\t\t\ttotalTokens += tokens;\n\t\t\t\t}\n\t\t\t}\n\t\t\t// Stop - we've hit the budget\n\t\t\tbreak;\n\t\t}\n\n\t\tmessages.unshift(message);\n\t\ttotalTokens += tokens;\n\t}\n\n\treturn { messages, fileOps, totalTokens };\n}\n\n// ============================================================================\n// Summary Generation\n// ============================================================================\n\nconst BRANCH_SUMMARY_PREAMBLE = `The user explored a different conversation branch before returning here.\nSummary of that exploration:\n\n`;\n\nconst BRANCH_SUMMARY_PROMPT = `Create a structured summary of this conversation branch for context when returning later.\n\nUse this EXACT format:\n\n## Goal\n[What was the user trying to accomplish in this branch?]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Work that was started but not finished]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [What should happen next to continue this work]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\n/**\n * Generate a summary of abandoned branch entries.\n *\n * @param entries - Session entries to summarize (chronological order)\n * @param options - Generation options\n */\nasync function completeBranchSummary(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n): Promise<AssistantMessage> {\n\tif (!streamFn) return completeSimple(model, context, options);\n\tconst stream = await streamFn(model, context, options);\n\treturn stream.result();\n}\n\nexport async function generateBranchSummary(\n\tentries: SessionEntry[],\n\toptions: GenerateBranchSummaryOptions,\n): Promise<BranchSummaryResult> {\n\tconst {\n\t\tmodel,\n\t\tapiKey,\n\t\theaders,\n\t\tsignal,\n\t\tcustomInstructions,\n\t\treplaceInstructions,\n\t\treserveTokens = 16384,\n\t\tstreamFn,\n\t} = options;\n\n\t// Token budget = context window minus reserved space for prompt + response\n\tconst contextWindow = model.contextWindow || 128000;\n\tconst tokenBudget = contextWindow - reserveTokens;\n\n\tconst { messages, fileOps } = prepareBranchEntries(entries, tokenBudget);\n\n\tif (messages.length === 0) {\n\t\treturn { summary: \"No content to summarize\" };\n\t}\n\n\t// Transform to LLM-compatible messages, then serialize to text\n\t// Serialization prevents the model from treating it as a conversation to continue\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build prompt\n\tlet instructions: string;\n\tif (replaceInstructions && customInstructions) {\n\t\tinstructions = customInstructions;\n\t} else if (customInstructions) {\n\t\tinstructions = `${BRANCH_SUMMARY_PROMPT}\\n\\nAdditional focus: ${customInstructions}`;\n\t} else {\n\t\tinstructions = BRANCH_SUMMARY_PROMPT;\n\t}\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${instructions}`;\n\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\t// Call LLM for summarization. Retry transient provider/watchdog failures so branch summaries\n\t// use the same reliability doctrine as compaction instead of bypassing the host stream wrapper.\n\tlet response: AssistantMessage | undefined;\n\tfor (let attempt = 0; attempt < 3; attempt++) {\n\t\tresponse = await completeBranchSummary(\n\t\t\tmodel,\n\t\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\t\t{ apiKey, headers, signal, maxTokens: 2048 },\n\t\t\tstreamFn,\n\t\t);\n\t\tif (response.stopReason !== \"error\") break;\n\t\tconst classified = classifyFailure({ message: response.errorMessage ?? \"\", provider: response.provider });\n\t\tif (!classified.retryable || signal.aborted) break;\n\t}\n\tif (!response) return { error: \"Summarization failed\" };\n\n\t// Check if aborted or errored\n\tif (response.stopReason === \"aborted\") {\n\t\treturn { aborted: true };\n\t}\n\tif (response.stopReason === \"error\") {\n\t\treturn { error: response.errorMessage || \"Summarization failed\" };\n\t}\n\n\tlet summary = response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n\n\t// Prepend preamble to provide context about the branch summary\n\tsummary = BRANCH_SUMMARY_PREAMBLE + summary;\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\treturn {\n\t\tsummary: summary || \"No summary generated\",\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}

package/dist/compaction/compaction.d.ts

@@ -78,7 +78,8 @@ export declare const MIN_COMPACTION_SAVINGS = 0.12;
 export declare function shouldCompact(contextTokens: number, contextWindow: number, settings: CompactionSettings, triggerTokens?: number): boolean;
 /**
  * Estimate token count for a message using chars/4 heuristic.
- * This is conservative (overestimates tokens).
+ * This is a rough planning heuristic; code and structured text can be denser than 4 chars/token,
+ * so callers that must stay under a provider bound need additional headroom.
  */
 export declare function estimateTokens(message: AgentMessage): number;
 /**
@@ -127,6 +128,9 @@ export declare const SUMMARY_BUDGET_MAX_TOKENS = 4000;
  * local servers silently truncate over-window prompts instead of erroring.
  */
 export declare function summarizerCanIngest(model: Model<any>, estimatedInputTokens: number): boolean;
+export declare function getChunkSummarizationTokenBudget(inputBound: number): number;
+export declare function buildChunkSummarizationPrompt(chunk: string, index: number, total: number): string;
+export declare function estimateStringTokens(text: string): number;
 export interface CompactionPreparation {
     /** UUID of first entry to keep */
     firstKeptEntryId: string;
@@ -146,7 +150,9 @@ export interface CompactionPreparation {
     /** Compaction settions from settings.jsonl	*/
     settings: CompactionSettings;
 }
-export declare function prepareCompaction(pathEntries: SessionEntry[], settings: CompactionSettings): CompactionPreparation | undefined;
+export declare function prepareCompaction(pathEntries: SessionEntry[], settings: CompactionSettings, options?: {
+    allowTrailingCompactionAsPrevious?: boolean;
+}): CompactionPreparation | undefined;
 /**
  * Generate summaries for compaction using prepared data.
  * Returns CompactionResult - SessionManager adds uuid/parentUuid when saving.
@@ -156,7 +162,6 @@ export declare function prepareCompaction(pathEntries: SessionEntry[], settings:
  */
 export declare function compact(preparation: CompactionPreparation, model: Model<any>, apiKey: string | undefined, headers?: Record<string, string>, customInstructions?: string, signal?: AbortSignal, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, preDigest?: (conversationText: string, signal?: AbortSignal) => Promise<string>, executionOptions?: {
     chunked?: boolean;
-    allowVerificationFailure?: boolean;
 }): Promise<CompactionResult>;
 export declare function createDeterministicCompaction(preparation: CompactionPreparation): CompactionResult;
 //# sourceMappingURL=compaction.d.ts.map

package/dist/compaction/compaction.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"compaction.d.ts","sourceRoot":"","sources":["../../src/compaction/compaction.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAA6B,KAAK,EAAuB,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAQtG,OAAO,EAA6C,KAAK,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAC7G,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AACzE,OAAO,EAAE,KAAK,eAAe,EAA4C,MAAM,iBAAiB,CAAC;AACjG,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAoB,KAAK,kBAAkB,EAAiB,MAAM,mBAAmB,CAAC;AAM7F,kEAAkE;AAClE,MAAM,WAAW,iBAAiB;IACjC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAkED,8EAA8E;AAC9E,MAAM,WAAW,gBAAgB,CAAC,CAAC,GAAG,OAAO;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,gBAAgB,EAAE,MAAM,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,+FAA+F;IAC/F,OAAO,CAAC,EAAE,CAAC,CAAC;IACZ,YAAY,CAAC,EAAE,kBAAkB,CAAC;CAClC;AAMD,MAAM,WAAW,kBAAkB;IAClC,OAAO,EAAE,OAAO,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB;;;;;;OAMG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,eAAO,MAAM,2BAA2B,EAAE,kBAKzC,CAAC;AAMF;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAgBD;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,YAAY,EAAE,GAAG,KAAK,GAAG,SAAS,CAShF;AAED,MAAM,WAAW,oBAAoB;IACpC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAUD;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,oBAAoB,CA4BpF;AAED;;;;;;GAMG;AACH,eAAO,MAAM,sBAAsB,OAAO,CAAC;AAE3C;;;;;;;;GAQG;AACH,wBAAgB,aAAa,CAC5B,aAAa,EAAE,MAAM,EACrB,aAAa,EAAE,MAAM,EACrB,QAAQ,EAAE,kBAAkB,EAC5B,aAAa,CAAC,EAAE,MAAM,GACpB,OAAO,CAmBT;AAwBD;;;GAGG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAwC5D;AAiDD;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAe1G;AAED,MAAM,WAAW,cAAc;IAC9B,mCAAmC;IACnC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,qFAAqF;IACrF,cAAc,EAAE,MAAM,CAAC;IACvB,uEAAuE;IACvE,WAAW,EAAE,OAAO,CAAC;CACrB;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,YAAY,CAC3B,OAAO,EAAE,YAAY,EAAE,EACvB,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,MAAM,EAChB,gBAAgB,EAAE,MAAM,GACtB,cAAc,CAyDhB;AAiED;;;GAGG;AACH,wBAAsB,eAAe,CACpC,eAAe,EAAE,YAAY,EAAE,EAC/B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,aAAa,EAAE,MAAM,EACrB,MAAM,EAAE,MAAM,GAAG,SAAS,EAC1B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAChC,MAAM,CAAC,EAAE,WAAW,EACpB,kBAAkB,CAAC,EAAE,MAAM,EAC3B,eAAe,CAAC,EAAE,MAAM,EACxB,aAAa,CAAC,EAAE,aAAa,EAC7B,QAAQ,CAAC,EAAE,QAAQ,EACnB,SAAS,CAAC,EAAE,CAAC,gBAAgB,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,MAAM,CAAC,EAC/E,UAAU,SAAoC,EAC9C,OAAO,UAAQ,GACb,OAAO,CAAC,MAAM,CAAC,CAqEjB;AAOD,mGAAmG;AACnG,eAAO,MAAM,yBAAyB,OAAQ,CAAC;AAqB/C;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,oBAAoB,EAAE,MAAM,GAAG,OAAO,CAI5F;AA8HD,MAAM,WAAW,qBAAqB;IACrC,kCAAkC;IAClC,gBAAgB,EAAE,MAAM,CAAC;IACzB,qDAAqD;IACrD,mBAAmB,EAAE,YAAY,EAAE,CAAC;IACpC,2EAA2E;IAC3E,kBAAkB,EAAE,YAAY,EAAE,CAAC;IACnC,iEAAiE;IACjE,WAAW,EAAE,OAAO,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,6DAA6D;IAC7D,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,yDAAyD;IACzD,OAAO,EAAE,cAAc,CAAC;IACxB,sEAAsE;IACtE,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,8CAA8C;IAC9C,QAAQ,EAAE,kBAAkB,CAAC;CAC7B;AAED,wBAAgB,iBAAiB,CAChC,WAAW,EAAE,YAAY,EAAE,EAC3B,QAAQ,EAAE,kBAAkB,GAC1B,qBAAqB,GAAG,SAAS,CA2EnC;AAqBD;;;;;;GAMG;AACH,wBAAsB,OAAO,CAC5B,WAAW,EAAE,qBAAqB,EAClC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,MAAM,EAAE,MAAM,GAAG,SAAS,EAC1B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAChC,kBAAkB,CAAC,EAAE,MAAM,EAC3B,MAAM,CAAC,EAAE,WAAW,EACpB,aAAa,CAAC,EAAE,aAAa,EAC7B,QAAQ,CAAC,EAAE,QAAQ,EACnB,SAAS,CAAC,EAAE,CAAC,gBAAgB,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,MAAM,CAAC,EAC/E,gBAAgB,CAAC,EAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAC;IAAC,wBAAwB,CAAC,EAAE,OAAO,CAAA;CAAE,GAC1E,OAAO,CAAC,gBAAgB,CAAC,CAwH3B;AAMD,wBAAgB,6BAA6B,CAAC,WAAW,EAAE,qBAAqB,GAAG,gBAAgB,CAwDlG","sourcesContent":["/**\n * Context compaction for long sessions.\n *\n * Pure functions for compaction logic. The session manager handles I/O,\n * and after compaction the session is reloaded.\n */\n\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions, Usage } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport { buildSessionContext, type CompactionEntry, type SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage, StreamFn, ThinkingLevel } from \"../types.ts\";\nimport { type CompactionFacts, extractCompactionFacts, renderFactsBlock } from \"./extraction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\nimport { buildRetryPrompt, type VerificationReport, verifySummary } from \"./verification.ts\";\n\n// ============================================================================\n// File Operation Tracking\n// ============================================================================\n\n/** Details stored in CompactionEntry.details for file tracking */\nexport interface CompactionDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\n/**\n * Extract file operations from messages and previous compaction entries.\n */\nfunction extractFileOperations(\n\tmessages: AgentMessage[],\n\tentries: SessionEntry[],\n\tprevCompactionIndex: number,\n): FileOperations {\n\tconst fileOps = createFileOps();\n\n\t// Collect from previous compaction's details (if pi-generated)\n\tif (prevCompactionIndex >= 0) {\n\t\tconst prevCompaction = entries[prevCompactionIndex] as CompactionEntry;\n\t\tif (!prevCompaction.fromHook && prevCompaction.details) {\n\t\t\t// fromHook field kept for session file compatibility\n\t\t\tconst details = prevCompaction.details as CompactionDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\tfor (const f of details.modifiedFiles) fileOps.edited.add(f);\n\t\t\t}\n\t\t}\n\t}\n\n\t// Extract from tool calls in messages\n\tfor (const msg of messages) {\n\t\textractFileOpsFromMessage(msg, fileOps);\n\t}\n\n\treturn fileOps;\n}\n\n// ============================================================================\n// Message Extraction\n// ============================================================================\n\n/**\n * Extract AgentMessage from an entry if it produces one.\n * Returns undefined for entries that don't contribute to LLM context.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"message\") {\n\t\treturn entry.message;\n\t}\n\tif (entry.type === \"custom_message\") {\n\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\t}\n\tif (entry.type === \"branch_summary\") {\n\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\t}\n\tif (entry.type === \"compaction\") {\n\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\t}\n\treturn undefined;\n}\n\nfunction getMessageFromEntryForCompaction(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"compaction\") {\n\t\treturn undefined;\n\t}\n\treturn getMessageFromEntry(entry);\n}\n\n/** Result from compact() - SessionManager adds uuid/parentUuid when saving */\nexport interface CompactionResult<T = unknown> {\n\tsummary: string;\n\tfirstKeptEntryId: string;\n\ttokensBefore: number;\n\t/** Extension-specific data (e.g., ArtifactIndex, version markers for structured compaction) */\n\tdetails?: T;\n\tverification?: VerificationReport;\n}\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface CompactionSettings {\n\tenabled: boolean;\n\treserveTokens: number;\n\tkeepRecentTokens: number;\n\t/**\n\t * Compaction also triggers once context exceeds this fraction of the model's window — not only when\n\t * it's nearly full (`contextWindow - reserveTokens`). On large-window models, waiting until nearly\n\t * full means every turn pays a huge input cost; a fractional cap keeps per-turn input bounded\n\t * (cost guard). The effective trigger is the LOWER of the two, so small-window models keep the\n\t * reserve-based behavior while large windows compact earlier. `0`/`1`+ disables the fractional cap.\n\t */\n\ttriggerPercent?: number;\n}\n\nexport const DEFAULT_COMPACTION_SETTINGS: CompactionSettings = {\n\tenabled: true,\n\treserveTokens: 16384,\n\tkeepRecentTokens: 20000,\n\ttriggerPercent: 0.7,\n};\n\n// ============================================================================\n// Token calculation\n// ============================================================================\n\n/**\n * Calculate total context tokens from usage.\n * Uses the native totalTokens field when available, falls back to computing from components.\n */\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\n/**\n * Get usage from an assistant message if available.\n * Skips aborted and error messages as they don't have valid usage data.\n */\nfunction getAssistantUsage(msg: AgentMessage): Usage | undefined {\n\tif (msg.role === \"assistant\" && \"usage\" in msg) {\n\t\tconst assistantMsg = msg as AssistantMessage;\n\t\tif (assistantMsg.stopReason !== \"aborted\" && assistantMsg.stopReason !== \"error\" && assistantMsg.usage) {\n\t\t\treturn assistantMsg.usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\n/**\n * Find the last non-aborted assistant message usage from session entries.\n */\nexport function getLastAssistantUsage(entries: SessionEntry[]): Usage | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"message\") {\n\t\t\tconst usage = getAssistantUsage(entry.message);\n\t\t\tif (usage) return usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\nexport interface ContextUsageEstimate {\n\ttokens: number;\n\tusageTokens: number;\n\ttrailingTokens: number;\n\tlastUsageIndex: number | null;\n}\n\nfunction getLastAssistantUsageInfo(messages: AgentMessage[]): { usage: Usage; index: number } | undefined {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst usage = getAssistantUsage(messages[i]);\n\t\tif (usage) return { usage, index: i };\n\t}\n\treturn undefined;\n}\n\n/**\n * Estimate context tokens from messages, using the last assistant usage when available.\n * If there are messages after the last usage, estimate their tokens with estimateTokens.\n */\nexport function estimateContextTokens(messages: AgentMessage[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\n\tif (!usageInfo) {\n\t\tlet estimated = 0;\n\t\tfor (const message of messages) {\n\t\t\testimated += estimateTokens(message);\n\t\t}\n\t\treturn {\n\t\t\ttokens: estimated,\n\t\t\tusageTokens: 0,\n\t\t\ttrailingTokens: estimated,\n\t\t\tlastUsageIndex: null,\n\t\t};\n\t}\n\n\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\tlet trailingTokens = 0;\n\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\ttrailingTokens += estimateTokens(messages[i]);\n\t}\n\n\treturn {\n\t\ttokens: usageTokens + trailingTokens,\n\t\tusageTokens,\n\t\ttrailingTokens,\n\t\tlastUsageIndex: usageInfo.index,\n\t};\n}\n\n/**\n * Minimum projected space saving for the EARLY (fractional) compaction trigger to fire. Anti-thrashing\n * (cost guard, #30): an early compaction whose summary would barely shrink the context (mostly recent,\n * protected content) just burns a summarization call for little gain — skip it and let the context grow\n * until either the saving is worthwhile or the hard (near-full) trigger forces it. Does NOT gate the\n * hard trigger, so overflow is always avoided.\n */\nexport const MIN_COMPACTION_SAVINGS = 0.12;\n\n/**\n * Check if compaction should trigger based on context usage.\n *\n * Two triggers:\n * - HARD: context exceeds `contextWindow - reserveTokens` (near-full) or an explicit `triggerTokens`\n *   override — always compact (prevents overflow).\n * - EARLY (fractional, cost guard): context exceeds `contextWindow * triggerPercent` — compact only if\n *   the summary would actually save enough (`MIN_COMPACTION_SAVINGS`), so we don't thrash for tiny gains.\n */\nexport function shouldCompact(\n\tcontextTokens: number,\n\tcontextWindow: number,\n\tsettings: CompactionSettings,\n\ttriggerTokens?: number,\n): boolean {\n\tif (!settings.enabled) return false;\n\n\t// Hard trigger: near-full, or a caller-supplied lower override. Always compacts (avoid overflow).\n\tconst reserveTrigger = contextWindow - settings.reserveTokens;\n\tconst hardTrigger = triggerTokens === undefined ? reserveTrigger : Math.min(reserveTrigger, triggerTokens);\n\tif (contextTokens > hardTrigger) return true;\n\n\t// Early fractional trigger: bounds per-turn input cost on large-window models, gated by anti-thrashing.\n\tconst pct = settings.triggerPercent ?? 0;\n\tif (pct > 0 && pct < 1) {\n\t\tconst fractionalTrigger = Math.floor(contextWindow * pct);\n\t\tif (contextTokens > fractionalTrigger) {\n\t\t\t// Projected saving ≈ the non-protected fraction (everything but the recent tail we keep).\n\t\t\tconst projectedSavings = contextTokens > 0 ? 1 - settings.keepRecentTokens / contextTokens : 0;\n\t\t\treturn projectedSavings >= MIN_COMPACTION_SAVINGS;\n\t\t}\n\t}\n\treturn false;\n}\n\n// ============================================================================\n// Cut point detection\n// ============================================================================\n\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nfunction estimateTextAndImageContentChars(content: string | Array<{ type: string; text?: string }>): number {\n\tif (typeof content === \"string\") {\n\t\treturn content.length;\n\t}\n\n\tlet chars = 0;\n\tfor (const block of content) {\n\t\tif (block.type === \"text\" && block.text) {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"image\") {\n\t\t\tchars += ESTIMATED_IMAGE_CHARS;\n\t\t}\n\t}\n\treturn chars;\n}\n\n/**\n * Estimate token count for a message using chars/4 heuristic.\n * This is conservative (overestimates tokens).\n */\nexport function estimateTokens(message: AgentMessage): number {\n\tlet chars = 0;\n\n\tswitch (message.role) {\n\t\tcase \"user\": {\n\t\t\tchars = estimateTextAndImageContentChars(\n\t\t\t\t(message as { content: string | Array<{ type: string; text?: string }> }).content,\n\t\t\t);\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"assistant\": {\n\t\t\tconst assistant = message as AssistantMessage;\n\t\t\tfor (const block of assistant.content) {\n\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\tchars += block.text.length;\n\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\tchars += block.thinking.length;\n\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\tchars += block.name.length + JSON.stringify(block.arguments).length;\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"custom\":\n\t\tcase \"toolResult\": {\n\t\t\tchars = estimateTextAndImageContentChars(message.content);\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"bashExecution\": {\n\t\t\tchars = message.command.length + message.output.length;\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\": {\n\t\t\tchars = message.summary.length;\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t}\n\n\treturn 0;\n}\n\n/**\n * Find valid cut points: indices of user, assistant, custom, or bashExecution messages.\n * Never cut at tool results (they must follow their tool call).\n * When we cut at an assistant message with tool calls, its tool results follow it\n * and will be kept.\n * BashExecutionMessage is treated like a user message (user-initiated context).\n */\nfunction findValidCutPoints(entries: SessionEntry[], startIndex: number, endIndex: number): number[] {\n\tconst cutPoints: number[] = [];\n\tfor (let i = startIndex; i < endIndex; i++) {\n\t\tconst entry = entries[i];\n\t\tswitch (entry.type) {\n\t\t\tcase \"message\": {\n\t\t\t\tconst role = entry.message.role;\n\t\t\t\tswitch (role) {\n\t\t\t\t\tcase \"bashExecution\":\n\t\t\t\t\tcase \"custom\":\n\t\t\t\t\tcase \"branchSummary\":\n\t\t\t\t\tcase \"compactionSummary\":\n\t\t\t\t\tcase \"user\":\n\t\t\t\t\tcase \"assistant\":\n\t\t\t\t\t\tcutPoints.push(i);\n\t\t\t\t\t\tbreak;\n\t\t\t\t\tcase \"toolResult\":\n\t\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"thinking_level_change\":\n\t\t\tcase \"model_change\":\n\t\t\tcase \"compaction\":\n\t\t\tcase \"branch_summary\":\n\t\t\tcase \"custom\":\n\t\t\tcase \"custom_message\":\n\t\t\tcase \"label\":\n\t\t\tcase \"session_info\":\n\t\t\t\tbreak;\n\t\t}\n\n\t\t// branch_summary and custom_message are user-role messages, valid cut points\n\t\tif (entry.type === \"branch_summary\" || entry.type === \"custom_message\") {\n\t\t\tcutPoints.push(i);\n\t\t}\n\t}\n\treturn cutPoints;\n}\n\n/**\n * Find the user message (or bashExecution) that starts the turn containing the given entry index.\n * Returns -1 if no turn start found before the index.\n * BashExecutionMessage is treated like a user message for turn boundaries.\n */\nexport function findTurnStartIndex(entries: SessionEntry[], entryIndex: number, startIndex: number): number {\n\tfor (let i = entryIndex; i >= startIndex; i--) {\n\t\tconst entry = entries[i];\n\t\t// branch_summary and custom_message are user-role messages, can start a turn\n\t\tif (entry.type === \"branch_summary\" || entry.type === \"custom_message\") {\n\t\t\treturn i;\n\t\t}\n\t\tif (entry.type === \"message\") {\n\t\t\tconst role = entry.message.role;\n\t\t\tif (role === \"user\" || role === \"bashExecution\") {\n\t\t\t\treturn i;\n\t\t\t}\n\t\t}\n\t}\n\treturn -1;\n}\n\nexport interface CutPointResult {\n\t/** Index of first entry to keep */\n\tfirstKeptEntryIndex: number;\n\t/** Index of user message that starts the turn being split, or -1 if not splitting */\n\tturnStartIndex: number;\n\t/** Whether this cut splits a turn (cut point is not a user message) */\n\tisSplitTurn: boolean;\n}\n\n/**\n * Find the cut point in session entries that keeps approximately `keepRecentTokens`.\n *\n * Algorithm: Walk backwards from newest, accumulating estimated message sizes.\n * Stop when we've accumulated >= keepRecentTokens. Cut at that point.\n *\n * Can cut at user OR assistant messages (never tool results). When cutting at an\n * assistant message with tool calls, its tool results come after and will be kept.\n *\n * Returns CutPointResult with:\n * - firstKeptEntryIndex: the entry index to start keeping from\n * - turnStartIndex: if cutting mid-turn, the user message that started that turn\n * - isSplitTurn: whether we're cutting in the middle of a turn\n *\n * Only considers entries between `startIndex` and `endIndex` (exclusive).\n */\nexport function findCutPoint(\n\tentries: SessionEntry[],\n\tstartIndex: number,\n\tendIndex: number,\n\tkeepRecentTokens: number,\n): CutPointResult {\n\tconst cutPoints = findValidCutPoints(entries, startIndex, endIndex);\n\n\tif (cutPoints.length === 0) {\n\t\treturn { firstKeptEntryIndex: startIndex, turnStartIndex: -1, isSplitTurn: false };\n\t}\n\n\t// Walk backwards from newest, accumulating estimated message sizes\n\tlet accumulatedTokens = 0;\n\tlet cutIndex = cutPoints[0]; // Default: keep from first message (not header)\n\n\tfor (let i = endIndex - 1; i >= startIndex; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type !== \"message\") continue;\n\n\t\t// Estimate this message's size\n\t\tconst messageTokens = estimateTokens(entry.message);\n\t\taccumulatedTokens += messageTokens;\n\n\t\t// Check if we've exceeded the budget\n\t\tif (accumulatedTokens >= keepRecentTokens) {\n\t\t\t// Find the closest valid cut point at or after this entry\n\t\t\tfor (let c = 0; c < cutPoints.length; c++) {\n\t\t\t\tif (cutPoints[c] >= i) {\n\t\t\t\t\tcutIndex = cutPoints[c];\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Scan backwards from cutIndex to include any non-message entries (bash, settings, etc.)\n\twhile (cutIndex > startIndex) {\n\t\tconst prevEntry = entries[cutIndex - 1];\n\t\t// Stop at session header or compaction boundaries\n\t\tif (prevEntry.type === \"compaction\") {\n\t\t\tbreak;\n\t\t}\n\t\tif (prevEntry.type === \"message\") {\n\t\t\t// Stop if we hit any message\n\t\t\tbreak;\n\t\t}\n\t\t// Include this non-message entry (bash, settings change, etc.)\n\t\tcutIndex--;\n\t}\n\n\t// Determine if this is a split turn\n\tconst cutEntry = entries[cutIndex];\n\tconst isUserMessage = cutEntry.type === \"message\" && cutEntry.message.role === \"user\";\n\tconst turnStartIndex = isUserMessage ? -1 : findTurnStartIndex(entries, cutIndex, startIndex);\n\n\treturn {\n\t\tfirstKeptEntryIndex: cutIndex,\n\t\tturnStartIndex,\n\t\tisSplitTurn: !isUserMessage && turnStartIndex !== -1,\n\t};\n}\n\n// ============================================================================\n// Summarization\n// ============================================================================\n\nconst SUMMARIZATION_PROMPT = `Checkpoint the conversation above. Format from your instructions, sections in this order:\n## Active Task\n### Mandatory Rules\n## Files\n## Done\n## Constraints & Preferences\n## Key Decisions\n## Blocked / Open\n## Critical Context\n\nPre-extracted facts (verified by tooling — include ALL of them, merged with what you read):\n<facts>\n{FACTS_BLOCK}\n</facts>\n\nBudget: ~{BUDGET} tokens. Concrete beats complete.`;\n\nconst UPDATE_SUMMARIZATION_PROMPT = `Update the checkpoint in <previous-summary> with the NEW turns above. RULES:\n- PRESERVE every existing ### Mandatory Rules bullet VERBATIM; append new ones.\n- Continue the ## Done numbering. Keep the 15 most recent numbered items verbatim; compress everything older into the single first line \"1. (earlier work compressed) <one line>\". The checkpoint must not grow without bound across updates.\n- Update ## Active Task to the newest unfulfilled user input; apply the cancellation rule.\n- Keep ## Files current (add new, keep still-relevant, drop obsolete).\n- Preserve exact paths, commands, errors.\n\nSame section order. Pre-extracted facts:\n<facts>\n{FACTS_BLOCK}\n</facts>\n\nBudget: ~{BUDGET} tokens.`;\n\nfunction createSummarizationOptions(\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n): SimpleStreamOptions {\n\tconst options: SimpleStreamOptions = { maxTokens, signal, apiKey, headers };\n\tif (model.reasoning && thinkingLevel && thinkingLevel !== \"off\") {\n\t\toptions.reasoning = thinkingLevel;\n\t}\n\treturn options;\n}\n\nasync function completeSummarization(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n): Promise<AssistantMessage> {\n\tif (!streamFn) {\n\t\treturn completeSimple(model, context, options);\n\t}\n\tconst stream = await streamFn(model, context, options);\n\treturn stream.result();\n}\n\n/**\n * Generate a summary of the conversation using the LLM.\n * If previousSummary is provided, uses the update prompt to merge.\n */\nexport async function generateSummary(\n\tcurrentMessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tcustomInstructions?: string,\n\tpreviousSummary?: string,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tpreDigest?: (conversationText: string, signal?: AbortSignal) => Promise<string>,\n\tfactsBlock = \"files:\\nactions:\\nprohibitions:\",\n\tchunked = false,\n): Promise<string> {\n\tconst summaryBudget = getSummaryBudget(reserveTokens, model, factsBlock);\n\tconst maxTokens = summaryBudget;\n\n\tlet promptSuffix = fillPromptTemplate(\n\t\tpreviousSummary ? UPDATE_SUMMARIZATION_PROMPT : SUMMARIZATION_PROMPT,\n\t\tfactsBlock,\n\t\tsummaryBudget,\n\t);\n\tif (customInstructions) {\n\t\tpromptSuffix = `${promptSuffix}\\n\\nAdditional focus: ${customInstructions}`;\n\t}\n\n\tconst llmMessages = convertToLlm(currentMessages);\n\tlet conversationText = serializeConversation(llmMessages);\n\tif (preDigest) {\n\t\ttry {\n\t\t\tconversationText = await preDigest(conversationText, signal);\n\t\t} catch {\n\t\t\t// Keep the verbatim conversation when an optional pre-digest fails.\n\t\t}\n\t}\n\n\tconst inputBound = getSummarizerInputBound(model, maxTokens);\n\tconst initialPromptText = buildSummarizationPrompt(conversationText, previousSummary, promptSuffix);\n\tif (estimateStringTokens(initialPromptText) > inputBound) {\n\t\tif (!chunked) {\n\t\t\tthrow new Error(\"input-overflow: summarization request exceeds summarizer window\");\n\t\t}\n\t\tconversationText = await summarizeChunks(\n\t\t\tconversationText,\n\t\t\tmodel,\n\t\t\tmaxTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tinputBound,\n\t\t);\n\t}\n\n\tconst promptText = buildSummarizationPrompt(conversationText, previousSummary, promptSuffix);\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\t{\n\t\t\tsystemPrompt: SUMMARIZATION_SYSTEM_PROMPT,\n\t\t\tmessages: [\n\t\t\t\t{\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: [{ type: \"text\", text: promptText }],\n\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t},\n\t\t\t],\n\t\t},\n\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\tstreamFn,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\t// A length-stopped checkpoint silently lost its tail sections — gating it as if complete\n\t// guarantees a verification failure. Fail loudly so the compaction ladder escalates instead.\n\tif (response.stopReason === \"length\") {\n\t\tthrow new Error(\"summary-length-stop: summarizer hit its output cap before completing the checkpoint\");\n\t}\n\n\treturn truncateSummaryToBudget(extractTextContent(response), summaryBudget);\n}\n\nfunction fillPromptTemplate(template: string, factsBlock: string, budget: number): string {\n\treturn template.replaceAll(\"{FACTS_BLOCK}\", factsBlock).replaceAll(\"{BUDGET}\", String(budget));\n}\n\nconst SUMMARY_BUDGET_BASE_TOKENS = 1_500;\n/** Hard ceiling for the facts-scaled summary budget; also the worst case selection must assume. */\nexport const SUMMARY_BUDGET_MAX_TOKENS = 4_000;\n/** Prompt-side margin beyond the raw conversation input (system prompt, tags, instructions). */\nconst SUMMARIZER_PROMPT_MARGIN_TOKENS = 2_000;\n\nfunction getSummaryBudget(reserveTokens: number, model: Model<any>, factsBlock?: string): number {\n\tconst modelMaxTokens = model.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY;\n\t// The verification gate demands the summary restate every extracted fact (modified files,\n\t// actions, rules). A fixed budget makes large spans structurally fail: the model length-stops\n\t// and the gated sections are the casualties. Scale the budget with the demand, bounded.\n\tconst factsTokens = factsBlock ? estimateStringTokens(factsBlock) : 0;\n\tconst demandBudget = Math.min(SUMMARY_BUDGET_MAX_TOKENS, Math.max(SUMMARY_BUDGET_BASE_TOKENS, factsTokens + 500));\n\treturn Math.max(1, Math.min(demandBudget, Math.floor(0.8 * reserveTokens), modelMaxTokens));\n}\n\nfunction getSummarizerInputBound(model: Model<any>, maxTokens: number): number {\n\tconst contextWindow = model.contextWindow > 0 ? model.contextWindow : Number.POSITIVE_INFINITY;\n\treturn contextWindow === Number.POSITIVE_INFINITY\n\t\t? contextWindow\n\t\t: Math.max(1, contextWindow - maxTokens - SUMMARIZER_PROMPT_MARGIN_TOKENS);\n}\n\n/**\n * Whether a candidate summarizer can ingest a summarization input of the given size in ONE\n * request (unchunked), using the same window arithmetic as {@link getSummarizerInputBound} with\n * the worst-case (facts-scaled) summary budget. Hosts use this at SELECTION time: a model that\n * fails this must not be handed the job — chunking cannot rescue recall-gated summarization, and\n * local servers silently truncate over-window prompts instead of erroring.\n */\nexport function summarizerCanIngest(model: Model<any>, estimatedInputTokens: number): boolean {\n\tconst contextWindow = model.contextWindow > 0 ? model.contextWindow : Number.POSITIVE_INFINITY;\n\tif (contextWindow === Number.POSITIVE_INFINITY) return true;\n\treturn estimatedInputTokens <= contextWindow - SUMMARY_BUDGET_MAX_TOKENS - SUMMARIZER_PROMPT_MARGIN_TOKENS;\n}\n\nfunction buildSummarizationPrompt(\n\tconversationText: string,\n\tpreviousSummary: string | undefined,\n\tpromptSuffix: string,\n): string {\n\tlet promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n`;\n\tif (previousSummary) {\n\t\tpromptText += `<previous-summary>\\n${previousSummary}\\n</previous-summary>\\n\\n`;\n\t}\n\treturn promptText + promptSuffix;\n}\n\nasync function summarizeChunks(\n\tconversationText: string,\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n\tstreamFn: StreamFn | undefined,\n\tinputBound: number,\n): Promise<string> {\n\tconst maxChunkTokens = Math.max(1, inputBound);\n\tconst maxChunkChars = Math.max(1, maxChunkTokens * 4);\n\tconst chunks = splitText(conversationText, maxChunkChars);\n\tconst retainedChunks = chunks.slice(-4);\n\tconst omittedChunks = chunks.length - retainedChunks.length;\n\tconst summaries: string[] = [];\n\n\tfor (let i = 0; i < retainedChunks.length; i++) {\n\t\tconst promptText = `<conversation-chunk index=\"${i + 1}\" total=\"${retainedChunks.length}\">\\n${retainedChunks[i]}\\n</conversation-chunk>\\n\\nSummarize this chunk for a later checkpoint merge. Preserve exact file paths, commands, errors, user prohibitions, and active work. Output concise notes only.`;\n\t\tconst response = await completeSummarization(\n\t\t\tmodel,\n\t\t\t{\n\t\t\t\tsystemPrompt: SUMMARIZATION_SYSTEM_PROMPT,\n\t\t\t\tmessages: [\n\t\t\t\t\t{\n\t\t\t\t\t\trole: \"user\",\n\t\t\t\t\t\tcontent: [{ type: \"text\", text: promptText }],\n\t\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t\t},\n\t\t\t\t],\n\t\t\t},\n\t\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\t\tstreamFn,\n\t\t);\n\t\tif (response.stopReason === \"error\") {\n\t\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t\t}\n\t\tsummaries.push(extractTextContent(response));\n\t}\n\n\tconst omittedNote =\n\t\tomittedChunks > 0\n\t\t\t? `## Critical Context\\n${omittedChunks} older oversized chunks omitted; deterministic facts supplied separately.\\n\\n`\n\t\t\t: \"\";\n\treturn `${omittedNote}${summaries.join(\"\\n\\n\")}`;\n}\n\nfunction splitText(text: string, maxChars: number): string[] {\n\tconst chunks: string[] = [];\n\tfor (let start = 0; start < text.length; start += maxChars) {\n\t\tchunks.push(text.slice(start, start + maxChars));\n\t}\n\treturn chunks.length > 0 ? chunks : [\"\"];\n}\n\nfunction extractTextContent(message: AssistantMessage): string {\n\treturn message.content\n\t\t.filter((content): content is { type: \"text\"; text: string } => content.type === \"text\")\n\t\t.map((content) => content.text)\n\t\t.join(\"\\n\");\n}\n\nfunction truncateSummaryToBudget(summary: string, budget: number): string {\n\tconst maxTokens = Math.floor(budget * 1.3);\n\tif (estimateStringTokens(summary) <= maxTokens) {\n\t\treturn summary;\n\t}\n\n\tlet current = summary;\n\t// Never drop \"Files\" or \"Done\" here: the verification gate checks exactly those sections\n\t// (files-modified/read-recall, actions-overlap), so deleting them guarantees gate failure.\n\tfor (const heading of [\"Critical Context\", \"Blocked / Open\", \"Key Decisions\", \"Constraints & Preferences\"]) {\n\t\tconst next = removeSummarySection(current, heading);\n\t\tif (next === current) {\n\t\t\tcontinue;\n\t\t}\n\t\tcurrent = next;\n\t\tif (estimateStringTokens(current) <= maxTokens) {\n\t\t\treturn current;\n\t\t}\n\t}\n\treturn current;\n}\n\nfunction removeSummarySection(summary: string, heading: string): string {\n\tconst lines = summary.split(/\\r?\\n/);\n\tconst kept: string[] = [];\n\tlet skipping = false;\n\tfor (const line of lines) {\n\t\tconst match = /^(?:##|###)\\s+(.+?)\\s*$/.exec(line);\n\t\tif (match) {\n\t\t\tskipping = match[1].trim().toLowerCase() === heading.toLowerCase();\n\t\t\tif (skipping) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t}\n\t\tif (!skipping) {\n\t\t\tkept.push(line);\n\t\t}\n\t}\n\treturn kept.join(\"\\n\").trim();\n}\n\nfunction estimateStringTokens(text: string): number {\n\treturn Math.ceil(text.length / 4);\n}\n\n// ============================================================================\n// Compaction Preparation (for extensions)\n// ============================================================================\n\nexport interface CompactionPreparation {\n\t/** UUID of first entry to keep */\n\tfirstKeptEntryId: string;\n\t/** Messages that will be summarized and discarded */\n\tmessagesToSummarize: AgentMessage[];\n\t/** Messages that will be turned into turn prefix summary (if splitting) */\n\tturnPrefixMessages: AgentMessage[];\n\t/** Whether this is a split turn (cut point in middle of turn) */\n\tisSplitTurn: boolean;\n\ttokensBefore: number;\n\t/** Summary from previous compaction, for iterative update */\n\tpreviousSummary?: string;\n\t/** File operations extracted from messagesToSummarize */\n\tfileOps: FileOperations;\n\t/** Facts extracted from the compacted span for verification gating */\n\tfacts?: CompactionFacts;\n\t/** Compaction settions from settings.jsonl\t*/\n\tsettings: CompactionSettings;\n}\n\nexport function prepareCompaction(\n\tpathEntries: SessionEntry[],\n\tsettings: CompactionSettings,\n): CompactionPreparation | undefined {\n\tif (pathEntries.length > 0 && pathEntries[pathEntries.length - 1].type === \"compaction\") {\n\t\treturn undefined;\n\t}\n\n\tlet prevCompactionIndex = -1;\n\tfor (let i = pathEntries.length - 1; i >= 0; i--) {\n\t\tif (pathEntries[i].type === \"compaction\") {\n\t\t\tprevCompactionIndex = i;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\tlet previousSummary: string | undefined;\n\tlet boundaryStart = 0;\n\tif (prevCompactionIndex >= 0) {\n\t\tconst prevCompaction = pathEntries[prevCompactionIndex] as CompactionEntry;\n\t\tpreviousSummary = prevCompaction.summary;\n\t\tconst firstKeptEntryIndex = pathEntries.findIndex((entry) => entry.id === prevCompaction.firstKeptEntryId);\n\t\tboundaryStart = firstKeptEntryIndex >= 0 ? firstKeptEntryIndex : prevCompactionIndex + 1;\n\t}\n\tconst boundaryEnd = pathEntries.length;\n\n\tconst tokensBefore = estimateContextTokens(buildSessionContext(pathEntries).messages).tokens;\n\n\tconst cutPoint = findCutPoint(pathEntries, boundaryStart, boundaryEnd, settings.keepRecentTokens);\n\n\t// Get UUID of first kept entry\n\tconst firstKeptEntry = pathEntries[cutPoint.firstKeptEntryIndex];\n\tif (!firstKeptEntry?.id) {\n\t\treturn undefined; // Session needs migration\n\t}\n\tconst firstKeptEntryId = firstKeptEntry.id;\n\n\tconst historyEnd = cutPoint.isSplitTurn ? cutPoint.turnStartIndex : cutPoint.firstKeptEntryIndex;\n\n\t// Messages to summarize (will be discarded after summary)\n\tconst messagesToSummarize: AgentMessage[] = [];\n\tfor (let i = boundaryStart; i < historyEnd; i++) {\n\t\tconst msg = getMessageFromEntryForCompaction(pathEntries[i]);\n\t\tif (msg) messagesToSummarize.push(msg);\n\t}\n\n\t// Messages for turn prefix summary (if splitting a turn)\n\tconst turnPrefixMessages: AgentMessage[] = [];\n\tif (cutPoint.isSplitTurn) {\n\t\tfor (let i = cutPoint.turnStartIndex; i < cutPoint.firstKeptEntryIndex; i++) {\n\t\t\tconst msg = getMessageFromEntryForCompaction(pathEntries[i]);\n\t\t\tif (msg) turnPrefixMessages.push(msg);\n\t\t}\n\t}\n\n\t// Extract file operations from messages and previous compaction\n\tconst fileOps = extractFileOperations(messagesToSummarize, pathEntries, prevCompactionIndex);\n\n\t// Also extract file ops from turn prefix if splitting\n\tif (cutPoint.isSplitTurn) {\n\t\tfor (const msg of turnPrefixMessages) {\n\t\t\textractFileOpsFromMessage(msg, fileOps);\n\t\t}\n\t}\n\n\tconst facts = extractCompactionFacts(pathEntries, boundaryStart, historyEnd);\n\n\treturn {\n\t\tfirstKeptEntryId,\n\t\tmessagesToSummarize,\n\t\tturnPrefixMessages,\n\t\tisSplitTurn: cutPoint.isSplitTurn,\n\t\ttokensBefore,\n\t\tpreviousSummary,\n\t\tfileOps,\n\t\tfacts,\n\t\tsettings,\n\t};\n}\n\n// ============================================================================\n// Main compaction function\n// ============================================================================\n\nconst TURN_PREFIX_SUMMARIZATION_PROMPT = `This is the PREFIX of a turn that was too large to keep. The SUFFIX (recent work) is retained.\n\nSummarize the prefix to provide context for the retained suffix:\n\n## Original Request\n[What did the user ask for in this turn?]\n\n## Early Progress\n- [Key decisions and work done in the prefix]\n\n## Context for Suffix\n- [Information needed to understand the retained recent work]\n\nBe concise. Focus on what's needed to understand the kept suffix.`;\n\n/**\n * Generate summaries for compaction using prepared data.\n * Returns CompactionResult - SessionManager adds uuid/parentUuid when saving.\n *\n * @param preparation - Pre-calculated preparation from prepareCompaction()\n * @param customInstructions - Optional custom focus for the summary\n */\nexport async function compact(\n\tpreparation: CompactionPreparation,\n\tmodel: Model<any>,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tcustomInstructions?: string,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tpreDigest?: (conversationText: string, signal?: AbortSignal) => Promise<string>,\n\texecutionOptions?: { chunked?: boolean; allowVerificationFailure?: boolean },\n): Promise<CompactionResult> {\n\tconst {\n\t\tfirstKeptEntryId,\n\t\tmessagesToSummarize,\n\t\tturnPrefixMessages,\n\t\tisSplitTurn,\n\t\ttokensBefore,\n\t\tpreviousSummary,\n\t\tfileOps,\n\t\tsettings,\n\t\tfacts: factsFromPreparation,\n\t} = preparation;\n\n\tconst facts = factsFromPreparation ?? {\n\t\tfiles: [],\n\t\tactions: [],\n\t\tprohibitions: [],\n\t\tcancelledText: \"\",\n\t\tactiveTaskSource: \"\",\n\t};\n\tconst factsBlock = renderFactsBlock(facts);\n\tlet verification: VerificationReport | undefined;\n\tlet summary = \"\";\n\n\tif (isSplitTurn && messagesToSummarize.length > 0) {\n\t\tlet historySummary = \"No prior history.\";\n\t\tlet historyInstructions = customInstructions;\n\t\tfor (let attempt = 0; attempt < 2; attempt++) {\n\t\t\thistorySummary = await generateSummary(\n\t\t\t\tmessagesToSummarize,\n\t\t\t\tmodel,\n\t\t\t\tsettings.reserveTokens,\n\t\t\t\tapiKey,\n\t\t\t\theaders,\n\t\t\t\tsignal,\n\t\t\t\thistoryInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tpreDigest,\n\t\t\t\tfactsBlock,\n\t\t\t\texecutionOptions?.chunked ?? false,\n\t\t\t);\n\n\t\t\tverification = verifySummary(historySummary, facts);\n\t\t\tif (verification.ok) {\n\t\t\t\tbreak;\n\t\t\t}\n\n\t\t\tif (attempt >= 1) {\n\t\t\t\tif (executionOptions?.allowVerificationFailure) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tthrow new Error(`gate-failed: ${formatVerificationFailures(verification)}`);\n\t\t\t}\n\n\t\t\thistoryInstructions = buildRetryPrompt(verification, historySummary);\n\t\t}\n\n\t\tconst turnPrefixSummary = await generateTurnPrefixSummary(\n\t\t\tturnPrefixMessages,\n\t\t\tmodel,\n\t\t\tsettings.reserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t);\n\t\tsummary = `${historySummary}\\n\\n---\\n\\n**Turn Context (split turn):**\\n\\n${turnPrefixSummary}`;\n\t} else {\n\t\tlet customSummaryInstructions = customInstructions;\n\t\tfor (let attempt = 0; attempt < 2; attempt++) {\n\t\t\tsummary = await generateSummary(\n\t\t\t\tmessagesToSummarize,\n\t\t\t\tmodel,\n\t\t\t\tsettings.reserveTokens,\n\t\t\t\tapiKey,\n\t\t\t\theaders,\n\t\t\t\tsignal,\n\t\t\t\tcustomSummaryInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tpreDigest,\n\t\t\t\tfactsBlock,\n\t\t\t\texecutionOptions?.chunked ?? false,\n\t\t\t);\n\n\t\t\tverification = verifySummary(summary, facts);\n\t\t\tif (verification.ok) {\n\t\t\t\tbreak;\n\t\t\t}\n\n\t\t\tif (attempt >= 1) {\n\t\t\t\tif (executionOptions?.allowVerificationFailure) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tthrow new Error(`gate-failed: ${formatVerificationFailures(verification)}`);\n\t\t\t}\n\n\t\t\tcustomSummaryInstructions = buildRetryPrompt(verification, summary);\n\t\t}\n\t}\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t\tverification,\n\t};\n}\n\nfunction formatVerificationFailures(verification: VerificationReport): string {\n\treturn verification.failures.map((failure) => `${failure.check}: ${failure.detail}`).join(\", \");\n}\n\nexport function createDeterministicCompaction(preparation: CompactionPreparation): CompactionResult {\n\tconst { firstKeptEntryId, tokensBefore, fileOps, facts } = preparation;\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tconst factsText = renderFactsBlock(\n\t\tfacts ?? {\n\t\t\tfiles: [],\n\t\t\tactions: [],\n\t\t\tprohibitions: [],\n\t\t\tcancelledText: \"\",\n\t\t\tactiveTaskSource: \"\",\n\t\t},\n\t);\n\tconst fileLines = facts?.files.length\n\t\t? facts.files.map((file) => `- ${file.path} — ${file.note || file.kind} (${file.kind})`)\n\t\t: [`- read: ${readFiles.length}`, `- modified: ${modifiedFiles.length}`];\n\tconst mandatoryRuleLines = facts?.prohibitions.length ? facts.prohibitions.map((rule) => `- ${rule}`) : [\"(none)\"];\n\tconst doneLines = facts?.actions.length\n\t\t? facts.actions.map((action, index) => `${index + 1}. ${action}`)\n\t\t: [\"1. CHECKPOINT deterministic fallback — repeated compaction retries exhausted\"];\n\tconst summary = [\n\t\t\"## Active Task\",\n\t\tfacts?.activeTaskSource ? `User: ${facts.activeTaskSource}` : \"Continue from the deterministic compact snapshot.\",\n\t\t\"\",\n\t\t\"### Mandatory Rules\",\n\t\t...mandatoryRuleLines,\n\t\t\"\",\n\t\t\"## Files\",\n\t\t...fileLines,\n\t\t\"\",\n\t\t\"## Done\",\n\t\t...doneLines,\n\t\t\"\",\n\t\t\"## Constraints & Preferences\",\n\t\t\"Preserve exact file paths, commands, line numbers, and error strings.\",\n\t\t\"\",\n\t\t\"## Key Decisions\",\n\t\t\"- Deterministic checkpoint used after repeated compaction retries.\",\n\t\t\"\",\n\t\t\"## Blocked / Open\",\n\t\t\"(none)\",\n\t\t\"\",\n\t\t\"## Critical Context\",\n\t\t\"- Deterministic facts-only checkpoint; no LLM summary was accepted.\",\n\t\tfactsText,\n\t].join(\"\\n\");\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t};\n}\n\n/**\n * Generate a summary for a turn prefix (when splitting a turn).\n */\nasync function generateTurnPrefixSummary(\n\tmessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n): Promise<string> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.5 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t); // Smaller budget for turn prefix\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${TURN_PREFIX_SUMMARIZATION_PROMPT}`;\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\tstreamFn,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Turn prefix summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\n\treturn response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"compaction.d.ts","sourceRoot":"","sources":["../../src/compaction/compaction.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAA6B,KAAK,EAAuB,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAQtG,OAAO,EAA6C,KAAK,YAAY,EAAE,MAAM,+BAA+B,CAAC;AAC7G,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AACzE,OAAO,EAAE,KAAK,eAAe,EAA4C,MAAM,iBAAiB,CAAC;AACjG,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAoB,KAAK,kBAAkB,EAAiB,MAAM,mBAAmB,CAAC;AAM7F,kEAAkE;AAClE,MAAM,WAAW,iBAAiB;IACjC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAkED,8EAA8E;AAC9E,MAAM,WAAW,gBAAgB,CAAC,CAAC,GAAG,OAAO;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,gBAAgB,EAAE,MAAM,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,+FAA+F;IAC/F,OAAO,CAAC,EAAE,CAAC,CAAC;IACZ,YAAY,CAAC,EAAE,kBAAkB,CAAC;CAClC;AAMD,MAAM,WAAW,kBAAkB;IAClC,OAAO,EAAE,OAAO,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;IACzB;;;;;;OAMG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,eAAO,MAAM,2BAA2B,EAAE,kBAKzC,CAAC;AAMF;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAgBD;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,YAAY,EAAE,GAAG,KAAK,GAAG,SAAS,CAShF;AAED,MAAM,WAAW,oBAAoB;IACpC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAC;CAC9B;AAUD;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,oBAAoB,CA4BpF;AAED;;;;;;GAMG;AACH,eAAO,MAAM,sBAAsB,OAAO,CAAC;AAE3C;;;;;;;;GAQG;AACH,wBAAgB,aAAa,CAC5B,aAAa,EAAE,MAAM,EACrB,aAAa,EAAE,MAAM,EACrB,QAAQ,EAAE,kBAAkB,EAC5B,aAAa,CAAC,EAAE,MAAM,GACpB,OAAO,CAmBT;AAwBD;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAwC5D;AAiDD;;;;GAIG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAe1G;AAED,MAAM,WAAW,cAAc;IAC9B,mCAAmC;IACnC,mBAAmB,EAAE,MAAM,CAAC;IAC5B,qFAAqF;IACrF,cAAc,EAAE,MAAM,CAAC;IACvB,uEAAuE;IACvE,WAAW,EAAE,OAAO,CAAC;CACrB;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,YAAY,CAC3B,OAAO,EAAE,YAAY,EAAE,EACvB,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,MAAM,EAChB,gBAAgB,EAAE,MAAM,GACtB,cAAc,CAyDhB;AAiED;;;GAGG;AACH,wBAAsB,eAAe,CACpC,eAAe,EAAE,YAAY,EAAE,EAC/B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,aAAa,EAAE,MAAM,EACrB,MAAM,EAAE,MAAM,GAAG,SAAS,EAC1B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAChC,MAAM,CAAC,EAAE,WAAW,EACpB,kBAAkB,CAAC,EAAE,MAAM,EAC3B,eAAe,CAAC,EAAE,MAAM,EACxB,aAAa,CAAC,EAAE,aAAa,EAC7B,QAAQ,CAAC,EAAE,QAAQ,EACnB,SAAS,CAAC,EAAE,CAAC,gBAAgB,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,MAAM,CAAC,EAC/E,UAAU,SAAoC,EAC9C,OAAO,UAAQ,GACb,OAAO,CAAC,MAAM,CAAC,CAqEjB;AAOD,mGAAmG;AACnG,eAAO,MAAM,yBAAyB,OAAQ,CAAC;AA2B/C;;;;;;GAMG;AACH,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EAAE,oBAAoB,EAAE,MAAM,GAAG,OAAO,CAI5F;AAgBD,wBAAgB,gCAAgC,CAAC,UAAU,EAAE,MAAM,GAAG,MAAM,CAE3E;AAED,wBAAgB,6BAA6B,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,MAAM,CAEjG;AA0GD,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEzD;AAMD,MAAM,WAAW,qBAAqB;IACrC,kCAAkC;IAClC,gBAAgB,EAAE,MAAM,CAAC;IACzB,qDAAqD;IACrD,mBAAmB,EAAE,YAAY,EAAE,CAAC;IACpC,2EAA2E;IAC3E,kBAAkB,EAAE,YAAY,EAAE,CAAC;IACnC,iEAAiE;IACjE,WAAW,EAAE,OAAO,CAAC;IACrB,YAAY,EAAE,MAAM,CAAC;IACrB,6DAA6D;IAC7D,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,yDAAyD;IACzD,OAAO,EAAE,cAAc,CAAC;IACxB,sEAAsE;IACtE,KAAK,CAAC,EAAE,eAAe,CAAC;IACxB,8CAA8C;IAC9C,QAAQ,EAAE,kBAAkB,CAAC;CAC7B;AAED,wBAAgB,iBAAiB,CAChC,WAAW,EAAE,YAAY,EAAE,EAC3B,QAAQ,EAAE,kBAAkB,EAC5B,OAAO,CAAC,EAAE;IAAE,iCAAiC,CAAC,EAAE,OAAO,CAAA;CAAE,GACvD,qBAAqB,GAAG,SAAS,CA+EnC;AAqBD;;;;;;GAMG;AACH,wBAAsB,OAAO,CAC5B,WAAW,EAAE,qBAAqB,EAClC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,MAAM,EAAE,MAAM,GAAG,SAAS,EAC1B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAChC,kBAAkB,CAAC,EAAE,MAAM,EAC3B,MAAM,CAAC,EAAE,WAAW,EACpB,aAAa,CAAC,EAAE,aAAa,EAC7B,QAAQ,CAAC,EAAE,QAAQ,EACnB,SAAS,CAAC,EAAE,CAAC,gBAAgB,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,KAAK,OAAO,CAAC,MAAM,CAAC,EAC/E,gBAAgB,CAAC,EAAE;IAAE,OAAO,CAAC,EAAE,OAAO,CAAA;CAAE,GACtC,OAAO,CAAC,gBAAgB,CAAC,CAkH3B;AAMD,wBAAgB,6BAA6B,CAAC,WAAW,EAAE,qBAAqB,GAAG,gBAAgB,CAwDlG","sourcesContent":["/**\n * Context compaction for long sessions.\n *\n * Pure functions for compaction logic. The session manager handles I/O,\n * and after compaction the session is reloaded.\n */\n\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions, Usage } from \"@caupulican/pi-ai\";\nimport { completeSimple } from \"@caupulican/pi-ai\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport { buildSessionContext, type CompactionEntry, type SessionEntry } from \"../session/session-manager.ts\";\nimport type { AgentMessage, StreamFn, ThinkingLevel } from \"../types.ts\";\nimport { type CompactionFacts, extractCompactionFacts, renderFactsBlock } from \"./extraction.ts\";\nimport {\n\tcomputeFileLists,\n\tcreateFileOps,\n\textractFileOpsFromMessage,\n\ttype FileOperations,\n\tformatFileOperations,\n\tSUMMARIZATION_SYSTEM_PROMPT,\n\tserializeConversation,\n} from \"./utils.ts\";\nimport { buildRetryPrompt, type VerificationReport, verifySummary } from \"./verification.ts\";\n\n// ============================================================================\n// File Operation Tracking\n// ============================================================================\n\n/** Details stored in CompactionEntry.details for file tracking */\nexport interface CompactionDetails {\n\treadFiles: string[];\n\tmodifiedFiles: string[];\n}\n\n/**\n * Extract file operations from messages and previous compaction entries.\n */\nfunction extractFileOperations(\n\tmessages: AgentMessage[],\n\tentries: SessionEntry[],\n\tprevCompactionIndex: number,\n): FileOperations {\n\tconst fileOps = createFileOps();\n\n\t// Collect from previous compaction's details (if pi-generated)\n\tif (prevCompactionIndex >= 0) {\n\t\tconst prevCompaction = entries[prevCompactionIndex] as CompactionEntry;\n\t\tif (!prevCompaction.fromHook && prevCompaction.details) {\n\t\t\t// fromHook field kept for session file compatibility\n\t\t\tconst details = prevCompaction.details as CompactionDetails;\n\t\t\tif (Array.isArray(details.readFiles)) {\n\t\t\t\tfor (const f of details.readFiles) fileOps.read.add(f);\n\t\t\t}\n\t\t\tif (Array.isArray(details.modifiedFiles)) {\n\t\t\t\tfor (const f of details.modifiedFiles) fileOps.edited.add(f);\n\t\t\t}\n\t\t}\n\t}\n\n\t// Extract from tool calls in messages\n\tfor (const msg of messages) {\n\t\textractFileOpsFromMessage(msg, fileOps);\n\t}\n\n\treturn fileOps;\n}\n\n// ============================================================================\n// Message Extraction\n// ============================================================================\n\n/**\n * Extract AgentMessage from an entry if it produces one.\n * Returns undefined for entries that don't contribute to LLM context.\n */\nfunction getMessageFromEntry(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"message\") {\n\t\treturn entry.message;\n\t}\n\tif (entry.type === \"custom_message\") {\n\t\treturn createCustomMessage(entry.customType, entry.content, entry.display, entry.details, entry.timestamp);\n\t}\n\tif (entry.type === \"branch_summary\") {\n\t\treturn createBranchSummaryMessage(entry.summary, entry.fromId, entry.timestamp);\n\t}\n\tif (entry.type === \"compaction\") {\n\t\treturn createCompactionSummaryMessage(entry.summary, entry.tokensBefore, entry.timestamp);\n\t}\n\treturn undefined;\n}\n\nfunction getMessageFromEntryForCompaction(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"compaction\") {\n\t\treturn undefined;\n\t}\n\treturn getMessageFromEntry(entry);\n}\n\n/** Result from compact() - SessionManager adds uuid/parentUuid when saving */\nexport interface CompactionResult<T = unknown> {\n\tsummary: string;\n\tfirstKeptEntryId: string;\n\ttokensBefore: number;\n\t/** Extension-specific data (e.g., ArtifactIndex, version markers for structured compaction) */\n\tdetails?: T;\n\tverification?: VerificationReport;\n}\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface CompactionSettings {\n\tenabled: boolean;\n\treserveTokens: number;\n\tkeepRecentTokens: number;\n\t/**\n\t * Compaction also triggers once context exceeds this fraction of the model's window — not only when\n\t * it's nearly full (`contextWindow - reserveTokens`). On large-window models, waiting until nearly\n\t * full means every turn pays a huge input cost; a fractional cap keeps per-turn input bounded\n\t * (cost guard). The effective trigger is the LOWER of the two, so small-window models keep the\n\t * reserve-based behavior while large windows compact earlier. `0`/`1`+ disables the fractional cap.\n\t */\n\ttriggerPercent?: number;\n}\n\nexport const DEFAULT_COMPACTION_SETTINGS: CompactionSettings = {\n\tenabled: true,\n\treserveTokens: 16384,\n\tkeepRecentTokens: 20000,\n\ttriggerPercent: 0.7,\n};\n\n// ============================================================================\n// Token calculation\n// ============================================================================\n\n/**\n * Calculate total context tokens from usage.\n * Uses the native totalTokens field when available, falls back to computing from components.\n */\nexport function calculateContextTokens(usage: Usage): number {\n\treturn usage.totalTokens || usage.input + usage.output + usage.cacheRead + usage.cacheWrite;\n}\n\n/**\n * Get usage from an assistant message if available.\n * Skips aborted and error messages as they don't have valid usage data.\n */\nfunction getAssistantUsage(msg: AgentMessage): Usage | undefined {\n\tif (msg.role === \"assistant\" && \"usage\" in msg) {\n\t\tconst assistantMsg = msg as AssistantMessage;\n\t\tif (assistantMsg.stopReason !== \"aborted\" && assistantMsg.stopReason !== \"error\" && assistantMsg.usage) {\n\t\t\treturn assistantMsg.usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\n/**\n * Find the last non-aborted assistant message usage from session entries.\n */\nexport function getLastAssistantUsage(entries: SessionEntry[]): Usage | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"message\") {\n\t\t\tconst usage = getAssistantUsage(entry.message);\n\t\t\tif (usage) return usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\nexport interface ContextUsageEstimate {\n\ttokens: number;\n\tusageTokens: number;\n\ttrailingTokens: number;\n\tlastUsageIndex: number | null;\n}\n\nfunction getLastAssistantUsageInfo(messages: AgentMessage[]): { usage: Usage; index: number } | undefined {\n\tfor (let i = messages.length - 1; i >= 0; i--) {\n\t\tconst usage = getAssistantUsage(messages[i]);\n\t\tif (usage) return { usage, index: i };\n\t}\n\treturn undefined;\n}\n\n/**\n * Estimate context tokens from messages, using the last assistant usage when available.\n * If there are messages after the last usage, estimate their tokens with estimateTokens.\n */\nexport function estimateContextTokens(messages: AgentMessage[]): ContextUsageEstimate {\n\tconst usageInfo = getLastAssistantUsageInfo(messages);\n\n\tif (!usageInfo) {\n\t\tlet estimated = 0;\n\t\tfor (const message of messages) {\n\t\t\testimated += estimateTokens(message);\n\t\t}\n\t\treturn {\n\t\t\ttokens: estimated,\n\t\t\tusageTokens: 0,\n\t\t\ttrailingTokens: estimated,\n\t\t\tlastUsageIndex: null,\n\t\t};\n\t}\n\n\tconst usageTokens = calculateContextTokens(usageInfo.usage);\n\tlet trailingTokens = 0;\n\tfor (let i = usageInfo.index + 1; i < messages.length; i++) {\n\t\ttrailingTokens += estimateTokens(messages[i]);\n\t}\n\n\treturn {\n\t\ttokens: usageTokens + trailingTokens,\n\t\tusageTokens,\n\t\ttrailingTokens,\n\t\tlastUsageIndex: usageInfo.index,\n\t};\n}\n\n/**\n * Minimum projected space saving for the EARLY (fractional) compaction trigger to fire. Anti-thrashing\n * (cost guard, #30): an early compaction whose summary would barely shrink the context (mostly recent,\n * protected content) just burns a summarization call for little gain — skip it and let the context grow\n * until either the saving is worthwhile or the hard (near-full) trigger forces it. Does NOT gate the\n * hard trigger, so overflow is always avoided.\n */\nexport const MIN_COMPACTION_SAVINGS = 0.12;\n\n/**\n * Check if compaction should trigger based on context usage.\n *\n * Two triggers:\n * - HARD: context exceeds `contextWindow - reserveTokens` (near-full) or an explicit `triggerTokens`\n *   override — always compact (prevents overflow).\n * - EARLY (fractional, cost guard): context exceeds `contextWindow * triggerPercent` — compact only if\n *   the summary would actually save enough (`MIN_COMPACTION_SAVINGS`), so we don't thrash for tiny gains.\n */\nexport function shouldCompact(\n\tcontextTokens: number,\n\tcontextWindow: number,\n\tsettings: CompactionSettings,\n\ttriggerTokens?: number,\n): boolean {\n\tif (!settings.enabled) return false;\n\n\t// Hard trigger: near-full, or a caller-supplied lower override. Always compacts (avoid overflow).\n\tconst reserveTrigger = contextWindow - settings.reserveTokens;\n\tconst hardTrigger = triggerTokens === undefined ? reserveTrigger : Math.min(reserveTrigger, triggerTokens);\n\tif (contextTokens > hardTrigger) return true;\n\n\t// Early fractional trigger: bounds per-turn input cost on large-window models, gated by anti-thrashing.\n\tconst pct = settings.triggerPercent ?? 0;\n\tif (pct > 0 && pct < 1) {\n\t\tconst fractionalTrigger = Math.floor(contextWindow * pct);\n\t\tif (contextTokens > fractionalTrigger) {\n\t\t\t// Projected saving ≈ the non-protected fraction (everything but the recent tail we keep).\n\t\t\tconst projectedSavings = contextTokens > 0 ? 1 - settings.keepRecentTokens / contextTokens : 0;\n\t\t\treturn projectedSavings >= MIN_COMPACTION_SAVINGS;\n\t\t}\n\t}\n\treturn false;\n}\n\n// ============================================================================\n// Cut point detection\n// ============================================================================\n\nconst ESTIMATED_IMAGE_CHARS = 4800;\n\nfunction estimateTextAndImageContentChars(content: string | Array<{ type: string; text?: string }>): number {\n\tif (typeof content === \"string\") {\n\t\treturn content.length;\n\t}\n\n\tlet chars = 0;\n\tfor (const block of content) {\n\t\tif (block.type === \"text\" && block.text) {\n\t\t\tchars += block.text.length;\n\t\t} else if (block.type === \"image\") {\n\t\t\tchars += ESTIMATED_IMAGE_CHARS;\n\t\t}\n\t}\n\treturn chars;\n}\n\n/**\n * Estimate token count for a message using chars/4 heuristic.\n * This is a rough planning heuristic; code and structured text can be denser than 4 chars/token,\n * so callers that must stay under a provider bound need additional headroom.\n */\nexport function estimateTokens(message: AgentMessage): number {\n\tlet chars = 0;\n\n\tswitch (message.role) {\n\t\tcase \"user\": {\n\t\t\tchars = estimateTextAndImageContentChars(\n\t\t\t\t(message as { content: string | Array<{ type: string; text?: string }> }).content,\n\t\t\t);\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"assistant\": {\n\t\t\tconst assistant = message as AssistantMessage;\n\t\t\tfor (const block of assistant.content) {\n\t\t\t\tif (block.type === \"text\") {\n\t\t\t\t\tchars += block.text.length;\n\t\t\t\t} else if (block.type === \"thinking\") {\n\t\t\t\t\tchars += block.thinking.length;\n\t\t\t\t} else if (block.type === \"toolCall\") {\n\t\t\t\t\tchars += block.name.length + JSON.stringify(block.arguments).length;\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"custom\":\n\t\tcase \"toolResult\": {\n\t\t\tchars = estimateTextAndImageContentChars(message.content);\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"bashExecution\": {\n\t\t\tchars = message.command.length + message.output.length;\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\": {\n\t\t\tchars = message.summary.length;\n\t\t\treturn Math.ceil(chars / 4);\n\t\t}\n\t}\n\n\treturn 0;\n}\n\n/**\n * Find valid cut points: indices of user, assistant, custom, or bashExecution messages.\n * Never cut at tool results (they must follow their tool call).\n * When we cut at an assistant message with tool calls, its tool results follow it\n * and will be kept.\n * BashExecutionMessage is treated like a user message (user-initiated context).\n */\nfunction findValidCutPoints(entries: SessionEntry[], startIndex: number, endIndex: number): number[] {\n\tconst cutPoints: number[] = [];\n\tfor (let i = startIndex; i < endIndex; i++) {\n\t\tconst entry = entries[i];\n\t\tswitch (entry.type) {\n\t\t\tcase \"message\": {\n\t\t\t\tconst role = entry.message.role;\n\t\t\t\tswitch (role) {\n\t\t\t\t\tcase \"bashExecution\":\n\t\t\t\t\tcase \"custom\":\n\t\t\t\t\tcase \"branchSummary\":\n\t\t\t\t\tcase \"compactionSummary\":\n\t\t\t\t\tcase \"user\":\n\t\t\t\t\tcase \"assistant\":\n\t\t\t\t\t\tcutPoints.push(i);\n\t\t\t\t\t\tbreak;\n\t\t\t\t\tcase \"toolResult\":\n\t\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"thinking_level_change\":\n\t\t\tcase \"model_change\":\n\t\t\tcase \"compaction\":\n\t\t\tcase \"branch_summary\":\n\t\t\tcase \"custom\":\n\t\t\tcase \"custom_message\":\n\t\t\tcase \"label\":\n\t\t\tcase \"session_info\":\n\t\t\t\tbreak;\n\t\t}\n\n\t\t// branch_summary and custom_message are user-role messages, valid cut points\n\t\tif (entry.type === \"branch_summary\" || entry.type === \"custom_message\") {\n\t\t\tcutPoints.push(i);\n\t\t}\n\t}\n\treturn cutPoints;\n}\n\n/**\n * Find the user message (or bashExecution) that starts the turn containing the given entry index.\n * Returns -1 if no turn start found before the index.\n * BashExecutionMessage is treated like a user message for turn boundaries.\n */\nexport function findTurnStartIndex(entries: SessionEntry[], entryIndex: number, startIndex: number): number {\n\tfor (let i = entryIndex; i >= startIndex; i--) {\n\t\tconst entry = entries[i];\n\t\t// branch_summary and custom_message are user-role messages, can start a turn\n\t\tif (entry.type === \"branch_summary\" || entry.type === \"custom_message\") {\n\t\t\treturn i;\n\t\t}\n\t\tif (entry.type === \"message\") {\n\t\t\tconst role = entry.message.role;\n\t\t\tif (role === \"user\" || role === \"bashExecution\") {\n\t\t\t\treturn i;\n\t\t\t}\n\t\t}\n\t}\n\treturn -1;\n}\n\nexport interface CutPointResult {\n\t/** Index of first entry to keep */\n\tfirstKeptEntryIndex: number;\n\t/** Index of user message that starts the turn being split, or -1 if not splitting */\n\tturnStartIndex: number;\n\t/** Whether this cut splits a turn (cut point is not a user message) */\n\tisSplitTurn: boolean;\n}\n\n/**\n * Find the cut point in session entries that keeps approximately `keepRecentTokens`.\n *\n * Algorithm: Walk backwards from newest, accumulating estimated message sizes.\n * Stop when we've accumulated >= keepRecentTokens. Cut at that point.\n *\n * Can cut at user OR assistant messages (never tool results). When cutting at an\n * assistant message with tool calls, its tool results come after and will be kept.\n *\n * Returns CutPointResult with:\n * - firstKeptEntryIndex: the entry index to start keeping from\n * - turnStartIndex: if cutting mid-turn, the user message that started that turn\n * - isSplitTurn: whether we're cutting in the middle of a turn\n *\n * Only considers entries between `startIndex` and `endIndex` (exclusive).\n */\nexport function findCutPoint(\n\tentries: SessionEntry[],\n\tstartIndex: number,\n\tendIndex: number,\n\tkeepRecentTokens: number,\n): CutPointResult {\n\tconst cutPoints = findValidCutPoints(entries, startIndex, endIndex);\n\n\tif (cutPoints.length === 0) {\n\t\treturn { firstKeptEntryIndex: startIndex, turnStartIndex: -1, isSplitTurn: false };\n\t}\n\n\t// Walk backwards from newest, accumulating estimated message sizes\n\tlet accumulatedTokens = 0;\n\tlet cutIndex = cutPoints[0]; // Default: keep from first message (not header)\n\n\tfor (let i = endIndex - 1; i >= startIndex; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type !== \"message\") continue;\n\n\t\t// Estimate this message's size\n\t\tconst messageTokens = estimateTokens(entry.message);\n\t\taccumulatedTokens += messageTokens;\n\n\t\t// Check if we've exceeded the budget\n\t\tif (accumulatedTokens >= keepRecentTokens) {\n\t\t\t// Find the closest valid cut point at or after this entry\n\t\t\tfor (let c = 0; c < cutPoints.length; c++) {\n\t\t\t\tif (cutPoints[c] >= i) {\n\t\t\t\t\tcutIndex = cutPoints[c];\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\t\t\tbreak;\n\t\t}\n\t}\n\n\t// Scan backwards from cutIndex to include any non-message entries (bash, settings, etc.)\n\twhile (cutIndex > startIndex) {\n\t\tconst prevEntry = entries[cutIndex - 1];\n\t\t// Stop at session header or compaction boundaries\n\t\tif (prevEntry.type === \"compaction\") {\n\t\t\tbreak;\n\t\t}\n\t\tif (prevEntry.type === \"message\") {\n\t\t\t// Stop if we hit any message\n\t\t\tbreak;\n\t\t}\n\t\t// Include this non-message entry (bash, settings change, etc.)\n\t\tcutIndex--;\n\t}\n\n\t// Determine if this is a split turn\n\tconst cutEntry = entries[cutIndex];\n\tconst isUserMessage = cutEntry.type === \"message\" && cutEntry.message.role === \"user\";\n\tconst turnStartIndex = isUserMessage ? -1 : findTurnStartIndex(entries, cutIndex, startIndex);\n\n\treturn {\n\t\tfirstKeptEntryIndex: cutIndex,\n\t\tturnStartIndex,\n\t\tisSplitTurn: !isUserMessage && turnStartIndex !== -1,\n\t};\n}\n\n// ============================================================================\n// Summarization\n// ============================================================================\n\nconst SUMMARIZATION_PROMPT = `Checkpoint the conversation above. Format from your instructions, sections in this order:\n## Active Task\n### Mandatory Rules\n## Files\n## Done\n## Constraints & Preferences\n## Key Decisions\n## Blocked / Open\n## Critical Context\n\nPre-extracted facts (verified by tooling — include ALL of them, merged with what you read):\n<facts>\n{FACTS_BLOCK}\n</facts>\n\nBudget: ~{BUDGET} tokens. Concrete beats complete.`;\n\nconst UPDATE_SUMMARIZATION_PROMPT = `Update the checkpoint in <previous-summary> with the NEW turns above. RULES:\n- PRESERVE every existing ### Mandatory Rules bullet VERBATIM; append new ones.\n- Continue the ## Done numbering. Keep the 15 most recent numbered items verbatim; compress everything older into the single first line \"1. (earlier work compressed) <one line>\". The checkpoint must not grow without bound across updates.\n- Update ## Active Task to the newest unfulfilled user input; apply the cancellation rule.\n- Keep ## Files current (add new, keep still-relevant, drop obsolete).\n- Preserve exact paths, commands, errors.\n\nSame section order. Pre-extracted facts:\n<facts>\n{FACTS_BLOCK}\n</facts>\n\nBudget: ~{BUDGET} tokens.`;\n\nfunction createSummarizationOptions(\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n): SimpleStreamOptions {\n\tconst options: SimpleStreamOptions = { maxTokens, signal, apiKey, headers };\n\tif (model.reasoning && thinkingLevel && thinkingLevel !== \"off\") {\n\t\toptions.reasoning = thinkingLevel;\n\t}\n\treturn options;\n}\n\nasync function completeSummarization(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n): Promise<AssistantMessage> {\n\tif (!streamFn) {\n\t\treturn completeSimple(model, context, options);\n\t}\n\tconst stream = await streamFn(model, context, options);\n\treturn stream.result();\n}\n\n/**\n * Generate a summary of the conversation using the LLM.\n * If previousSummary is provided, uses the update prompt to merge.\n */\nexport async function generateSummary(\n\tcurrentMessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tcustomInstructions?: string,\n\tpreviousSummary?: string,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tpreDigest?: (conversationText: string, signal?: AbortSignal) => Promise<string>,\n\tfactsBlock = \"files:\\nactions:\\nprohibitions:\",\n\tchunked = false,\n): Promise<string> {\n\tconst summaryBudget = getSummaryBudget(reserveTokens, model, factsBlock);\n\tconst maxTokens = summaryBudget;\n\n\tlet promptSuffix = fillPromptTemplate(\n\t\tpreviousSummary ? UPDATE_SUMMARIZATION_PROMPT : SUMMARIZATION_PROMPT,\n\t\tfactsBlock,\n\t\tsummaryBudget,\n\t);\n\tif (customInstructions) {\n\t\tpromptSuffix = `${promptSuffix}\\n\\nAdditional focus: ${customInstructions}`;\n\t}\n\n\tconst llmMessages = convertToLlm(currentMessages);\n\tlet conversationText = serializeConversation(llmMessages);\n\tif (preDigest) {\n\t\ttry {\n\t\t\tconversationText = await preDigest(conversationText, signal);\n\t\t} catch {\n\t\t\t// Keep the verbatim conversation when an optional pre-digest fails.\n\t\t}\n\t}\n\n\tconst inputBound = getSummarizerInputBound(model, maxTokens);\n\tconst initialPromptText = buildSummarizationPrompt(conversationText, previousSummary, promptSuffix);\n\tif (estimateStringTokens(initialPromptText) > inputBound) {\n\t\tif (!chunked) {\n\t\t\tthrow new Error(\"input-overflow: summarization request exceeds summarizer window\");\n\t\t}\n\t\tconversationText = await summarizeChunks(\n\t\t\tconversationText,\n\t\t\tmodel,\n\t\t\tmaxTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tinputBound,\n\t\t);\n\t}\n\n\tconst promptText = buildSummarizationPrompt(conversationText, previousSummary, promptSuffix);\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\t{\n\t\t\tsystemPrompt: SUMMARIZATION_SYSTEM_PROMPT,\n\t\t\tmessages: [\n\t\t\t\t{\n\t\t\t\t\trole: \"user\",\n\t\t\t\t\tcontent: [{ type: \"text\", text: promptText }],\n\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t},\n\t\t\t],\n\t\t},\n\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\tstreamFn,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\t// A length-stopped checkpoint silently lost its tail sections — gating it as if complete\n\t// guarantees a verification failure. Fail loudly so the compaction ladder escalates instead.\n\tif (response.stopReason === \"length\") {\n\t\tthrow new Error(\"summary-length-stop: summarizer hit its output cap before completing the checkpoint\");\n\t}\n\n\treturn truncateSummaryToBudget(extractTextContent(response), summaryBudget);\n}\n\nfunction fillPromptTemplate(template: string, factsBlock: string, budget: number): string {\n\treturn template.replaceAll(\"{FACTS_BLOCK}\", factsBlock).replaceAll(\"{BUDGET}\", String(budget));\n}\n\nconst SUMMARY_BUDGET_BASE_TOKENS = 1_500;\n/** Hard ceiling for the facts-scaled summary budget; also the worst case selection must assume. */\nexport const SUMMARY_BUDGET_MAX_TOKENS = 4_000;\n/** Prompt-side margin beyond the raw conversation input (system prompt, tags, instructions). */\nconst SUMMARIZER_PROMPT_MARGIN_TOKENS = 2_000;\n\nfunction getSummaryBudget(reserveTokens: number, model: Model<any>, factsBlock?: string): number {\n\tconst modelMaxTokens = model.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY;\n\t// The verification gate demands the summary restate every extracted fact (modified files,\n\t// actions, rules). A fixed budget makes large spans structurally fail: the model length-stops\n\t// and the gated sections are the casualties. Scale the budget with the demand, bounded.\n\tconst factsTokens = factsBlock ? estimateStringTokens(factsBlock) : 0;\n\tconst demandBudget = Math.min(SUMMARY_BUDGET_MAX_TOKENS, Math.max(SUMMARY_BUDGET_BASE_TOKENS, factsTokens + 500));\n\treturn Math.max(1, Math.min(demandBudget, Math.floor(0.8 * reserveTokens), modelMaxTokens));\n}\n\nfunction getEffectiveContextWindow(model: Model<any>): number {\n\tconst registered = model.contextWindow > 0 ? model.contextWindow : Number.POSITIVE_INFINITY;\n\tconst served = (model as { servedContextWindow?: unknown }).servedContextWindow;\n\treturn typeof served === \"number\" && served > 0 ? Math.min(registered, served) : registered;\n}\n\nfunction getSummarizerInputBound(model: Model<any>, maxTokens: number): number {\n\tconst contextWindow = getEffectiveContextWindow(model);\n\treturn contextWindow === Number.POSITIVE_INFINITY\n\t\t? contextWindow\n\t\t: Math.max(1, contextWindow - maxTokens - SUMMARIZER_PROMPT_MARGIN_TOKENS);\n}\n\n/**\n * Whether a candidate summarizer can ingest a summarization input of the given size in ONE\n * request (unchunked), using the same window arithmetic as {@link getSummarizerInputBound} with\n * the worst-case (facts-scaled) summary budget. Hosts use this at SELECTION time: a model that\n * fails this must not be handed the job — chunking cannot rescue recall-gated summarization, and\n * local servers silently truncate over-window prompts instead of erroring.\n */\nexport function summarizerCanIngest(model: Model<any>, estimatedInputTokens: number): boolean {\n\tconst contextWindow = getEffectiveContextWindow(model);\n\tif (contextWindow === Number.POSITIVE_INFINITY) return true;\n\treturn estimatedInputTokens <= contextWindow - SUMMARY_BUDGET_MAX_TOKENS - SUMMARIZER_PROMPT_MARGIN_TOKENS;\n}\n\nfunction buildSummarizationPrompt(\n\tconversationText: string,\n\tpreviousSummary: string | undefined,\n\tpromptSuffix: string,\n): string {\n\tlet promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n`;\n\tif (previousSummary) {\n\t\tpromptText += `<previous-summary>\\n${previousSummary}\\n</previous-summary>\\n\\n`;\n\t}\n\treturn promptText + promptSuffix;\n}\n\nconst CHUNK_SUMMARIZATION_HEADROOM_TOKENS = 1000;\n\nexport function getChunkSummarizationTokenBudget(inputBound: number): number {\n\treturn Math.max(1, inputBound - CHUNK_SUMMARIZATION_HEADROOM_TOKENS);\n}\n\nexport function buildChunkSummarizationPrompt(chunk: string, index: number, total: number): string {\n\treturn `<conversation-chunk index=\"${index}\" total=\"${total}\">\\n${chunk}\\n</conversation-chunk>\\n\\nSummarize this chunk for a later checkpoint merge. Preserve exact file paths, commands, errors, user prohibitions, and active work. Output concise notes only.`;\n}\n\nasync function summarizeChunks(\n\tconversationText: string,\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n\tstreamFn: StreamFn | undefined,\n\tinputBound: number,\n): Promise<string> {\n\tconst maxChunkTokens = getChunkSummarizationTokenBudget(inputBound);\n\tconst maxChunkChars = Math.max(1, maxChunkTokens * 4);\n\tconst chunks = splitText(conversationText, maxChunkChars);\n\tconst retainedChunks = chunks.slice(-4);\n\tconst omittedChunks = chunks.length - retainedChunks.length;\n\tconst summaries: string[] = [];\n\n\tfor (let i = 0; i < retainedChunks.length; i++) {\n\t\tconst promptText = buildChunkSummarizationPrompt(retainedChunks[i], i + 1, retainedChunks.length);\n\t\tconst response = await completeSummarization(\n\t\t\tmodel,\n\t\t\t{\n\t\t\t\tsystemPrompt: SUMMARIZATION_SYSTEM_PROMPT,\n\t\t\t\tmessages: [\n\t\t\t\t\t{\n\t\t\t\t\t\trole: \"user\",\n\t\t\t\t\t\tcontent: [{ type: \"text\", text: promptText }],\n\t\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t\t},\n\t\t\t\t],\n\t\t\t},\n\t\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\t\tstreamFn,\n\t\t);\n\t\tif (response.stopReason === \"error\") {\n\t\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t\t}\n\t\tsummaries.push(extractTextContent(response));\n\t}\n\n\tconst omittedNote =\n\t\tomittedChunks > 0\n\t\t\t? `## Critical Context\\n${omittedChunks} older oversized chunks omitted; deterministic facts supplied separately.\\n\\n`\n\t\t\t: \"\";\n\treturn `${omittedNote}${summaries.join(\"\\n\\n\")}`;\n}\n\nfunction splitText(text: string, maxChars: number): string[] {\n\tconst chunks: string[] = [];\n\tfor (let start = 0; start < text.length; start += maxChars) {\n\t\tchunks.push(text.slice(start, start + maxChars));\n\t}\n\treturn chunks.length > 0 ? chunks : [\"\"];\n}\n\nfunction extractTextContent(message: AssistantMessage): string {\n\treturn message.content\n\t\t.filter((content): content is { type: \"text\"; text: string } => content.type === \"text\")\n\t\t.map((content) => content.text)\n\t\t.join(\"\\n\");\n}\n\nfunction truncateSummaryToBudget(summary: string, budget: number): string {\n\tconst maxTokens = Math.floor(budget * 1.3);\n\tif (estimateStringTokens(summary) <= maxTokens) {\n\t\treturn summary;\n\t}\n\n\tlet current = summary;\n\t// Never drop \"Files\" or \"Done\" here: the verification gate checks exactly those sections\n\t// (files-modified/read-recall, actions-overlap), so deleting them guarantees gate failure.\n\tfor (const heading of [\"Critical Context\", \"Blocked / Open\", \"Key Decisions\", \"Constraints & Preferences\"]) {\n\t\tconst next = removeSummarySection(current, heading);\n\t\tif (next === current) {\n\t\t\tcontinue;\n\t\t}\n\t\tcurrent = next;\n\t\tif (estimateStringTokens(current) <= maxTokens) {\n\t\t\treturn current;\n\t\t}\n\t}\n\treturn current;\n}\n\nfunction removeSummarySection(summary: string, heading: string): string {\n\tconst lines = summary.split(/\\r?\\n/);\n\tconst kept: string[] = [];\n\tlet skipping = false;\n\tfor (const line of lines) {\n\t\tconst match = /^(?:##|###)\\s+(.+?)\\s*$/.exec(line);\n\t\tif (match) {\n\t\t\tskipping = match[1].trim().toLowerCase() === heading.toLowerCase();\n\t\t\tif (skipping) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t}\n\t\tif (!skipping) {\n\t\t\tkept.push(line);\n\t\t}\n\t}\n\treturn kept.join(\"\\n\").trim();\n}\n\nexport function estimateStringTokens(text: string): number {\n\treturn Math.ceil(text.length / 4);\n}\n\n// ============================================================================\n// Compaction Preparation (for extensions)\n// ============================================================================\n\nexport interface CompactionPreparation {\n\t/** UUID of first entry to keep */\n\tfirstKeptEntryId: string;\n\t/** Messages that will be summarized and discarded */\n\tmessagesToSummarize: AgentMessage[];\n\t/** Messages that will be turned into turn prefix summary (if splitting) */\n\tturnPrefixMessages: AgentMessage[];\n\t/** Whether this is a split turn (cut point in middle of turn) */\n\tisSplitTurn: boolean;\n\ttokensBefore: number;\n\t/** Summary from previous compaction, for iterative update */\n\tpreviousSummary?: string;\n\t/** File operations extracted from messagesToSummarize */\n\tfileOps: FileOperations;\n\t/** Facts extracted from the compacted span for verification gating */\n\tfacts?: CompactionFacts;\n\t/** Compaction settions from settings.jsonl\t*/\n\tsettings: CompactionSettings;\n}\n\nexport function prepareCompaction(\n\tpathEntries: SessionEntry[],\n\tsettings: CompactionSettings,\n\toptions?: { allowTrailingCompactionAsPrevious?: boolean },\n): CompactionPreparation | undefined {\n\tconst trailingEntry = pathEntries[pathEntries.length - 1];\n\tif (trailingEntry?.type === \"compaction\" && !options?.allowTrailingCompactionAsPrevious) {\n\t\treturn undefined;\n\t}\n\n\tlet prevCompactionIndex = -1;\n\tfor (let i = pathEntries.length - 1; i >= 0; i--) {\n\t\tif (pathEntries[i].type === \"compaction\") {\n\t\t\tprevCompactionIndex = i;\n\t\t\tbreak;\n\t\t}\n\t}\n\n\tlet previousSummary: string | undefined;\n\tlet boundaryStart = 0;\n\tif (prevCompactionIndex >= 0) {\n\t\tconst prevCompaction = pathEntries[prevCompactionIndex] as CompactionEntry;\n\t\tpreviousSummary = prevCompaction.summary;\n\t\tconst firstKeptEntryIndex = pathEntries.findIndex((entry) => entry.id === prevCompaction.firstKeptEntryId);\n\t\tboundaryStart = firstKeptEntryIndex >= 0 ? firstKeptEntryIndex : prevCompactionIndex + 1;\n\t}\n\tconst boundaryEnd =\n\t\toptions?.allowTrailingCompactionAsPrevious && pathEntries[pathEntries.length - 1]?.type === \"compaction\"\n\t\t\t? pathEntries.length - 1\n\t\t\t: pathEntries.length;\n\n\tconst tokensBefore = estimateContextTokens(buildSessionContext(pathEntries).messages).tokens;\n\n\tconst cutPoint = findCutPoint(pathEntries, boundaryStart, boundaryEnd, settings.keepRecentTokens);\n\n\t// Get UUID of first kept entry\n\tconst firstKeptEntry = pathEntries[cutPoint.firstKeptEntryIndex];\n\tif (!firstKeptEntry?.id) {\n\t\treturn undefined; // Session needs migration\n\t}\n\tconst firstKeptEntryId = firstKeptEntry.id;\n\n\tconst historyEnd = cutPoint.isSplitTurn ? cutPoint.turnStartIndex : cutPoint.firstKeptEntryIndex;\n\n\t// Messages to summarize (will be discarded after summary)\n\tconst messagesToSummarize: AgentMessage[] = [];\n\tfor (let i = boundaryStart; i < historyEnd; i++) {\n\t\tconst msg = getMessageFromEntryForCompaction(pathEntries[i]);\n\t\tif (msg) messagesToSummarize.push(msg);\n\t}\n\n\t// Messages for turn prefix summary (if splitting a turn)\n\tconst turnPrefixMessages: AgentMessage[] = [];\n\tif (cutPoint.isSplitTurn) {\n\t\tfor (let i = cutPoint.turnStartIndex; i < cutPoint.firstKeptEntryIndex; i++) {\n\t\t\tconst msg = getMessageFromEntryForCompaction(pathEntries[i]);\n\t\t\tif (msg) turnPrefixMessages.push(msg);\n\t\t}\n\t}\n\n\t// Extract file operations from messages and previous compaction\n\tconst fileOps = extractFileOperations(messagesToSummarize, pathEntries, prevCompactionIndex);\n\n\t// Also extract file ops from turn prefix if splitting\n\tif (cutPoint.isSplitTurn) {\n\t\tfor (const msg of turnPrefixMessages) {\n\t\t\textractFileOpsFromMessage(msg, fileOps);\n\t\t}\n\t}\n\n\tconst facts = extractCompactionFacts(pathEntries, boundaryStart, historyEnd);\n\n\treturn {\n\t\tfirstKeptEntryId,\n\t\tmessagesToSummarize,\n\t\tturnPrefixMessages,\n\t\tisSplitTurn: cutPoint.isSplitTurn,\n\t\ttokensBefore,\n\t\tpreviousSummary,\n\t\tfileOps,\n\t\tfacts,\n\t\tsettings,\n\t};\n}\n\n// ============================================================================\n// Main compaction function\n// ============================================================================\n\nconst TURN_PREFIX_SUMMARIZATION_PROMPT = `This is the PREFIX of a turn that was too large to keep. The SUFFIX (recent work) is retained.\n\nSummarize the prefix to provide context for the retained suffix:\n\n## Original Request\n[What did the user ask for in this turn?]\n\n## Early Progress\n- [Key decisions and work done in the prefix]\n\n## Context for Suffix\n- [Information needed to understand the retained recent work]\n\nBe concise. Focus on what's needed to understand the kept suffix.`;\n\n/**\n * Generate summaries for compaction using prepared data.\n * Returns CompactionResult - SessionManager adds uuid/parentUuid when saving.\n *\n * @param preparation - Pre-calculated preparation from prepareCompaction()\n * @param customInstructions - Optional custom focus for the summary\n */\nexport async function compact(\n\tpreparation: CompactionPreparation,\n\tmodel: Model<any>,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tcustomInstructions?: string,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tpreDigest?: (conversationText: string, signal?: AbortSignal) => Promise<string>,\n\texecutionOptions?: { chunked?: boolean },\n): Promise<CompactionResult> {\n\tconst {\n\t\tfirstKeptEntryId,\n\t\tmessagesToSummarize,\n\t\tturnPrefixMessages,\n\t\tisSplitTurn,\n\t\ttokensBefore,\n\t\tpreviousSummary,\n\t\tfileOps,\n\t\tsettings,\n\t\tfacts: factsFromPreparation,\n\t} = preparation;\n\n\tconst facts = factsFromPreparation ?? {\n\t\tfiles: [],\n\t\tactions: [],\n\t\tprohibitions: [],\n\t\tcancelledText: \"\",\n\t\tactiveTaskSource: \"\",\n\t};\n\tconst factsBlock = renderFactsBlock(facts);\n\tlet verification: VerificationReport | undefined;\n\tlet summary = \"\";\n\n\tif (isSplitTurn && messagesToSummarize.length > 0) {\n\t\tlet historySummary = \"No prior history.\";\n\t\tlet historyInstructions = customInstructions;\n\t\tfor (let attempt = 0; attempt < 2; attempt++) {\n\t\t\thistorySummary = await generateSummary(\n\t\t\t\tmessagesToSummarize,\n\t\t\t\tmodel,\n\t\t\t\tsettings.reserveTokens,\n\t\t\t\tapiKey,\n\t\t\t\theaders,\n\t\t\t\tsignal,\n\t\t\t\thistoryInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tpreDigest,\n\t\t\t\tfactsBlock,\n\t\t\t\texecutionOptions?.chunked ?? false,\n\t\t\t);\n\n\t\t\tverification = verifySummary(historySummary, facts);\n\t\t\tif (verification.ok) {\n\t\t\t\tbreak;\n\t\t\t}\n\n\t\t\tif (attempt >= 1) {\n\t\t\t\tthrow new Error(`gate-failed: ${formatVerificationFailures(verification)}`);\n\t\t\t}\n\n\t\t\thistoryInstructions = buildRetryPrompt(verification, historySummary);\n\t\t}\n\n\t\tconst turnPrefixSummary = await generateTurnPrefixSummary(\n\t\t\tturnPrefixMessages,\n\t\t\tmodel,\n\t\t\tsettings.reserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t);\n\t\tsummary = `${historySummary}\\n\\n---\\n\\n**Turn Context (split turn):**\\n\\n${turnPrefixSummary}`;\n\t} else {\n\t\tlet customSummaryInstructions = customInstructions;\n\t\tfor (let attempt = 0; attempt < 2; attempt++) {\n\t\t\tsummary = await generateSummary(\n\t\t\t\tmessagesToSummarize,\n\t\t\t\tmodel,\n\t\t\t\tsettings.reserveTokens,\n\t\t\t\tapiKey,\n\t\t\t\theaders,\n\t\t\t\tsignal,\n\t\t\t\tcustomSummaryInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tpreDigest,\n\t\t\t\tfactsBlock,\n\t\t\t\texecutionOptions?.chunked ?? false,\n\t\t\t);\n\n\t\t\tverification = verifySummary(summary, facts);\n\t\t\tif (verification.ok) {\n\t\t\t\tbreak;\n\t\t\t}\n\n\t\t\tif (attempt >= 1) {\n\t\t\t\tthrow new Error(`gate-failed: ${formatVerificationFailures(verification)}`);\n\t\t\t}\n\n\t\t\tcustomSummaryInstructions = buildRetryPrompt(verification, summary);\n\t\t}\n\t}\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t\tverification,\n\t};\n}\n\nfunction formatVerificationFailures(verification: VerificationReport): string {\n\treturn verification.failures.map((failure) => `${failure.check}: ${failure.detail}`).join(\", \");\n}\n\nexport function createDeterministicCompaction(preparation: CompactionPreparation): CompactionResult {\n\tconst { firstKeptEntryId, tokensBefore, fileOps, facts } = preparation;\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tconst factsText = renderFactsBlock(\n\t\tfacts ?? {\n\t\t\tfiles: [],\n\t\t\tactions: [],\n\t\t\tprohibitions: [],\n\t\t\tcancelledText: \"\",\n\t\t\tactiveTaskSource: \"\",\n\t\t},\n\t);\n\tconst fileLines = facts?.files.length\n\t\t? facts.files.map((file) => `- ${file.path} — ${file.note || file.kind} (${file.kind})`)\n\t\t: [`- read: ${readFiles.length}`, `- modified: ${modifiedFiles.length}`];\n\tconst mandatoryRuleLines = facts?.prohibitions.length ? facts.prohibitions.map((rule) => `- ${rule}`) : [\"(none)\"];\n\tconst doneLines = facts?.actions.length\n\t\t? facts.actions.map((action, index) => `${index + 1}. ${action}`)\n\t\t: [\"1. CHECKPOINT deterministic fallback — repeated compaction retries exhausted\"];\n\tconst summary = [\n\t\t\"## Active Task\",\n\t\tfacts?.activeTaskSource ? `User: ${facts.activeTaskSource}` : \"Continue from the deterministic compact snapshot.\",\n\t\t\"\",\n\t\t\"### Mandatory Rules\",\n\t\t...mandatoryRuleLines,\n\t\t\"\",\n\t\t\"## Files\",\n\t\t...fileLines,\n\t\t\"\",\n\t\t\"## Done\",\n\t\t...doneLines,\n\t\t\"\",\n\t\t\"## Constraints & Preferences\",\n\t\t\"Preserve exact file paths, commands, line numbers, and error strings.\",\n\t\t\"\",\n\t\t\"## Key Decisions\",\n\t\t\"- Deterministic checkpoint used after repeated compaction retries.\",\n\t\t\"\",\n\t\t\"## Blocked / Open\",\n\t\t\"(none)\",\n\t\t\"\",\n\t\t\"## Critical Context\",\n\t\t\"- Deterministic facts-only checkpoint; no LLM summary was accepted.\",\n\t\tfactsText,\n\t].join(\"\\n\");\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t};\n}\n\n/**\n * Generate a summary for a turn prefix (when splitting a turn).\n */\nasync function generateTurnPrefixSummary(\n\tmessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n): Promise<string> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.5 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t); // Smaller budget for turn prefix\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${TURN_PREFIX_SUMMARIZATION_PROMPT}`;\n\tconst summarizationMessages = [\n\t\t{\n\t\t\trole: \"user\" as const,\n\t\t\tcontent: [{ type: \"text\" as const, text: promptText }],\n\t\t\ttimestamp: Date.now(),\n\t\t},\n\t];\n\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, signal, thinkingLevel),\n\t\tstreamFn,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Turn prefix summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\n\treturn response.content\n\t\t.filter((c): c is { type: \"text\"; text: string } => c.type === \"text\")\n\t\t.map((c) => c.text)\n\t\t.join(\"\\n\");\n}\n"]}