apex-code 0.0.1-alpha.7 → 0.0.1-alpha.9

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

@@ -1 +1 @@
-{"version":3,"file":"branch-summarization.js","sourceRoot":"","sources":["../../../src/core/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAGpD,OAAO,EACN,YAAY,EACZ,0BAA0B,EAC1B,8BAA8B,EAC9B,mBAAmB,GACnB,MAAM,gBAAgB,CAAC;AAExB,OAAO,EAAE,qBAAqB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACxE,OAAO,EACN,gBAAgB,EAChB,aAAa,EACb,yBAAyB,EAEzB,oBAAoB,EACpB,2BAA2B,EAC3B,qBAAqB,GACrB,MAAM,YAAY,CAAC;AAgEpB,+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,EACL,KAAK,EACL,MAAM,EACN,OAAO,EACP,GAAG,EACH,MAAM,EACN,kBAAkB,EAClB,mBAAmB,EACnB,aAAa,GAAG,KAAK,EACrB,QAAQ,EACR,KAAK,EACL,SAAS,GACT,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,wEAAwE;IACxE,6EAA6E;IAC7E,gFAAgF;IAChF,+DAA+D;IAC/D,MAAM,OAAO,GAAG,EAAE,YAAY,EAAE,2BAA2B,EAAE,QAAQ,EAAE,qBAAqB,EAAE,CAAC;IAC/F,MAAM,cAAc,GAAwB,EAAE,MAAM,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IAC9F,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAEzG,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,WAAW,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IAE5C,+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,KAAK,EAAE,QAAQ,CAAC,KAAK;QACrB,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 { RetryCallbacks, RetryPolicy } from \"@earendil-works/pi-ai\";\nimport { contentText } from \"@earendil-works/pi-ai\";\nimport type { Model, SimpleStreamOptions, Usage } from \"@earendil-works/pi-ai/compat\";\nimport type { AgentMessage, StreamFn } from \"apex-code-agent-core\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session-manager.ts\";\nimport { completeSummarization, 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\tusage?: Usage;\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/** Provider-scoped environment values for the model */\n\tenv?: 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 session stream function. Used to preserve SDK request behavior without mutating agent state. */\n\tstreamFn?: StreamFn;\n\t/** Retry policy for transient summarization errors. Reuses coding-agent's `settings.retry`. */\n\tretry?: RetryPolicy;\n\t/** Optional callbacks for retry reporting (e.g. TUI retry indicators). */\n\tcallbacks?: RetryCallbacks;\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 {\n\t\tmodel,\n\t\tapiKey,\n\t\theaders,\n\t\tenv,\n\t\tsignal,\n\t\tcustomInstructions,\n\t\treplaceInstructions,\n\t\treserveTokens = 16384,\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\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. Prefer the session stream function so SDK\n\t// request behavior (timeouts, retries, attribution headers) stays consistent\n\t// without running through agent state/events. Retried via completeSummarization\n\t// so transient stream drops reuse the configured retry policy.\n\tconst context = { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages };\n\tconst requestOptions: SimpleStreamOptions = { apiKey, headers, env, signal, maxTokens: 2048 };\n\tconst response = await completeSummarization(model, context, requestOptions, streamFn, retry, callbacks);\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 = contentText(response.content);\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\tusage: response.usage,\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}
+{"version":3,"file":"branch-summarization.js","sourceRoot":"","sources":["../../../src/core/compaction/branch-summarization.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAGpD,OAAO,EACN,YAAY,EACZ,0BAA0B,EAC1B,8BAA8B,EAC9B,mBAAmB,GACnB,MAAM,gBAAgB,CAAC;AAExB,OAAO,EAAE,qBAAqB,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AACxE,OAAO,EACN,gBAAgB,EAChB,aAAa,EACb,yBAAyB,EAEzB,oBAAoB,EACpB,2BAA2B,EAC3B,qBAAqB,GACrB,MAAM,YAAY,CAAC;AAgEpB,+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,EACL,KAAK,EACL,MAAM,EACN,OAAO,EACP,GAAG,EACH,MAAM,EACN,kBAAkB,EAClB,mBAAmB,EACnB,aAAa,GAAG,KAAK,EACrB,QAAQ,EACR,KAAK,EACL,SAAS,GACT,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,wEAAwE;IACxE,6EAA6E;IAC7E,gFAAgF;IAChF,+DAA+D;IAC/D,MAAM,OAAO,GAAG,EAAE,YAAY,EAAE,2BAA2B,EAAE,QAAQ,EAAE,qBAAqB,EAAE,CAAC;IAC/F,MAAM,cAAc,GAAwB,EAAE,MAAM,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IAC9F,MAAM,QAAQ,GAAG,MAAM,qBAAqB,CAAC,KAAK,EAAE,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;IAEzG,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;IACD,IAAI,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,UAAU,CAAC,EAAE,CAAC;QACjE,OAAO,EAAE,KAAK,EAAE,+CAA+C,EAAE,CAAC;IACnE,CAAC;IAED,IAAI,OAAO,GAAG,WAAW,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;IAE5C,+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,KAAK,EAAE,QAAQ,CAAC,KAAK;QACrB,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 { RetryCallbacks, RetryPolicy } from \"@earendil-works/pi-ai\";\nimport { contentText } from \"@earendil-works/pi-ai\";\nimport type { Model, SimpleStreamOptions, Usage } from \"@earendil-works/pi-ai/compat\";\nimport type { AgentMessage, StreamFn } from \"apex-code-agent-core\";\nimport {\n\tconvertToLlm,\n\tcreateBranchSummaryMessage,\n\tcreateCompactionSummaryMessage,\n\tcreateCustomMessage,\n} from \"../messages.ts\";\nimport type { ReadonlySessionManager, SessionEntry } from \"../session-manager.ts\";\nimport { completeSummarization, 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\tusage?: Usage;\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/** Provider-scoped environment values for the model */\n\tenv?: 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 session stream function. Used to preserve SDK request behavior without mutating agent state. */\n\tstreamFn?: StreamFn;\n\t/** Retry policy for transient summarization errors. Reuses coding-agent's `settings.retry`. */\n\tretry?: RetryPolicy;\n\t/** Optional callbacks for retry reporting (e.g. TUI retry indicators). */\n\tcallbacks?: RetryCallbacks;\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 {\n\t\tmodel,\n\t\tapiKey,\n\t\theaders,\n\t\tenv,\n\t\tsignal,\n\t\tcustomInstructions,\n\t\treplaceInstructions,\n\t\treserveTokens = 16384,\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\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. Prefer the session stream function so SDK\n\t// request behavior (timeouts, retries, attribution headers) stays consistent\n\t// without running through agent state/events. Retried via completeSummarization\n\t// so transient stream drops reuse the configured retry policy.\n\tconst context = { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages };\n\tconst requestOptions: SimpleStreamOptions = { apiKey, headers, env, signal, maxTokens: 2048 };\n\tconst response = await completeSummarization(model, context, requestOptions, streamFn, retry, callbacks);\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\tif (response.content.some((block) => block.type === \"toolCall\")) {\n\t\treturn { error: \"Branch summarization attempted to call a tool\" };\n\t}\n\n\tlet summary = contentText(response.content);\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\tusage: response.usage,\n\t\treadFiles,\n\t\tmodifiedFiles,\n\t};\n}\n"]}

package/dist/core/compaction/compaction.d.ts

@@ -102,9 +102,9 @@ export declare function completeSummarization(model: Model<any>, context: Contex
  * Generate a summary of the conversation using the LLM.
  * If previousSummary is provided, uses the update prompt to merge.
  */
-export declare function generateSummary(currentMessages: AgentMessage[], model: Model<any>, reserveTokens: number, apiKey: string | undefined, headers?: Record<string, string>, signal?: AbortSignal, customInstructions?: string, previousSummary?: string, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks): Promise<string>;
+export declare function generateSummary(currentMessages: AgentMessage[], model: Model<any>, reserveTokens: number, apiKey: string | undefined, headers?: Record<string, string>, signal?: AbortSignal, customInstructions?: string, previousSummary?: string, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks, sessionId?: string): Promise<string>;
 /** Generate or update a conversation summary and return its provider usage. */
-export declare function generateSummaryWithUsage(currentMessages: AgentMessage[], model: Model<any>, reserveTokens: number, apiKey: string | undefined, headers?: Record<string, string>, signal?: AbortSignal, customInstructions?: string, previousSummary?: string, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks): Promise<{
+export declare function generateSummaryWithUsage(currentMessages: AgentMessage[], model: Model<any>, reserveTokens: number, apiKey: string | undefined, headers?: Record<string, string>, signal?: AbortSignal, customInstructions?: string, previousSummary?: string, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks, sessionId?: string): Promise<{
     text: string;
     usage: Usage;
 }>;
@@ -132,6 +132,7 @@ export declare function prepareCompaction(pathEntries: SessionEntry[], settings:
  *
  * @param preparation - Pre-calculated preparation from prepareCompaction()
  * @param customInstructions - Optional custom focus for the summary
+ * @param sessionId - Optional routing session ID forwarded without enabling prompt caching
  */
-export declare function compact(preparation: CompactionPreparation, model: Model<any>, apiKey: string | undefined, headers?: Record<string, string>, customInstructions?: string, signal?: AbortSignal, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks): Promise<CompactionResult>;
+export declare function compact(preparation: CompactionPreparation, model: Model<any>, apiKey: string | undefined, headers?: Record<string, string>, customInstructions?: string, signal?: AbortSignal, thinkingLevel?: ThinkingLevel, streamFn?: StreamFn, env?: Record<string, string>, retry?: RetryPolicy, callbacks?: RetryCallbacks, sessionId?: string): Promise<CompactionResult>;
 //# sourceMappingURL=compaction.d.ts.map

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

@@ -1 +1 @@
-{"version":3,"file":"compaction.d.ts","sourceRoot":"","sources":["../../../src/core/compaction/compaction.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAe,KAAK,cAAc,EAAE,KAAK,WAAW,EAA8B,MAAM,uBAAuB,CAAC;AACvH,OAAO,KAAK,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,EAAE,MAAM,8BAA8B,CAAC;AAEjH,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAElF,OAAO,EAGN,KAAK,YAAY,EAEjB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AAMpB,kEAAkE;AAClE,MAAM,WAAW,iBAAiB;IACjC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAkDD,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,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,2EAA2E;IAC3E,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,+FAA+F;IAC/F,OAAO,CAAC,EAAE,CAAC,CAAC;CACZ;AA6BD,MAAM,WAAW,kBAAkB;IAClC,OAAO,EAAE,OAAO,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;CACzB;AAED,eAAO,MAAM,2BAA2B,EAAE,kBAIzC,CAAC;AAMF;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAqBD;;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;;GAEG;AACH,wBAAgB,aAAa,CAAC,aAAa,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,EAAE,QAAQ,EAAE,kBAAkB,GAAG,OAAO,CAGjH;AAwBD;;;GAGG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAwC5D;AA2DD;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAO1G;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,CAqDhB;AA8FD;;;;;;GAMG;AACH,wBAAsB,qBAAqB,CAC1C,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,mBAAmB,EAC5B,QAAQ,CAAC,EAAE,QAAQ,EACnB,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CAY3B;AAED;;;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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,MAAM,CAAC,CAkBjB;AAED,+EAA+E;AAC/E,wBAAsB,wBAAwB,CAC7C,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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,KAAK,CAAA;CAAE,CAAC,CAkDzC;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,8CAA8C;IAC9C,QAAQ,EAAE,kBAAkB,CAAC;CAC7B;AAED,wBAAgB,iBAAiB,CAChC,WAAW,EAAE,YAAY,EAAE,EAC3B,QAAQ,EAAE,kBAAkB,GAC1B,qBAAqB,GAAG,SAAS,CA4EnC;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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CA0F3B","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 { contentText, type RetryCallbacks, type RetryPolicy, retryAssistantCall, uuidv7 } from \"@earendil-works/pi-ai\";\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions, Usage } from \"@earendil-works/pi-ai/compat\";\nimport { completeSimple } from \"@earendil-works/pi-ai/compat\";\nimport type { AgentMessage, StreamFn, ThinkingLevel } from \"apex-code-agent-core\";\nimport { convertToLlm } from \"../messages.ts\";\nimport {\n\tbuildSessionContext,\n\ttype CompactionEntry,\n\ttype SessionEntry,\n\tsessionEntryToContextMessages,\n} from \"../session-manager.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// 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 getMessageFromEntryForCompaction(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"compaction\") {\n\t\treturn undefined;\n\t}\n\treturn sessionEntryToContextMessages(entry)[0];\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\testimatedTokensAfter?: number;\n\t/** Usage from the LLM call(s) that generated this summary, if available */\n\tusage?: Usage;\n\t/** Extension-specific data (e.g., ArtifactIndex, version markers for structured compaction) */\n\tdetails?: T;\n}\n\nfunction combineUsage(first: Usage, second: Usage): Usage {\n\treturn {\n\t\tinput: first.input + second.input,\n\t\toutput: first.output + second.output,\n\t\tcacheRead: first.cacheRead + second.cacheRead,\n\t\tcacheWrite: first.cacheWrite + second.cacheWrite,\n\t\t...(first.cacheWrite1h !== undefined || second.cacheWrite1h !== undefined\n\t\t\t? { cacheWrite1h: (first.cacheWrite1h ?? 0) + (second.cacheWrite1h ?? 0) }\n\t\t\t: {}),\n\t\t...(first.reasoning !== undefined || second.reasoning !== undefined\n\t\t\t? { reasoning: (first.reasoning ?? 0) + (second.reasoning ?? 0) }\n\t\t\t: {}),\n\t\ttotalTokens: first.totalTokens + second.totalTokens,\n\t\tcost: {\n\t\t\tinput: first.cost.input + second.cost.input,\n\t\t\toutput: first.cost.output + second.cost.output,\n\t\t\tcacheRead: first.cost.cacheRead + second.cost.cacheRead,\n\t\t\tcacheWrite: first.cost.cacheWrite + second.cost.cacheWrite,\n\t\t\ttotal: first.cost.total + second.cost.total,\n\t\t},\n\t};\n}\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface CompactionSettings {\n\tenabled: boolean;\n\treserveTokens: number;\n\tkeepRecentTokens: number;\n}\n\nexport const DEFAULT_COMPACTION_SETTINGS: CompactionSettings = {\n\tenabled: true,\n\treserveTokens: 16384,\n\tkeepRecentTokens: 20000,\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, error, and all-zero usage 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 (\n\t\t\tassistantMsg.stopReason !== \"aborted\" &&\n\t\t\tassistantMsg.stopReason !== \"error\" &&\n\t\t\tassistantMsg.usage &&\n\t\t\tcalculateContextTokens(assistantMsg.usage) > 0\n\t\t) {\n\t\t\treturn assistantMsg.usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\n/**\n * Find the last valid 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 * Check if compaction should trigger based on context usage.\n */\nexport function shouldCompact(contextTokens: number, contextWindow: number, settings: CompactionSettings): boolean {\n\tif (!settings.enabled) return false;\n\treturn contextTokens > contextWindow - settings.reserveTokens;\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\nfunction isCutPointMessage(message: AgentMessage): boolean {\n\tswitch (message.role) {\n\t\tcase \"user\":\n\t\tcase \"assistant\":\n\t\tcase \"bashExecution\":\n\t\tcase \"custom\":\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\":\n\t\t\treturn true;\n\t\tcase \"toolResult\":\n\t\t\treturn false;\n\t}\n\treturn false;\n}\n\nfunction isTurnStartMessage(message: AgentMessage): boolean {\n\tswitch (message.role) {\n\t\tcase \"user\":\n\t\tcase \"bashExecution\":\n\t\tcase \"custom\":\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\":\n\t\t\treturn true;\n\t\tcase \"assistant\":\n\t\tcase \"toolResult\":\n\t\t\treturn false;\n\t}\n\treturn false;\n}\n\nfunction isTurnStartEntry(entry: SessionEntry): boolean {\n\tif (entry.type === \"compaction\") {\n\t\treturn false;\n\t}\n\treturn sessionEntryToContextMessages(entry).some(isTurnStartMessage);\n}\n\n/**\n * Find valid cut points: indices of context-visible user-like or assistant 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 */\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\tif (entry.type === \"compaction\") {\n\t\t\tcontinue;\n\t\t}\n\t\tif (sessionEntryToContextMessages(entry).some(isCutPointMessage)) {\n\t\t\tcutPoints.push(i);\n\t\t}\n\t}\n\treturn cutPoints;\n}\n\n/**\n * Find the context-visible user-role message that starts the turn containing the given entry index.\n * Returns -1 if no turn start found before the index.\n */\nexport function findTurnStartIndex(entries: SessionEntry[], entryIndex: number, startIndex: number): number {\n\tfor (let i = entryIndex; i >= startIndex; i--) {\n\t\tif (isTurnStartEntry(entries[i])) {\n\t\t\treturn i;\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\tconst messageTokens = sessionEntryToContextMessages(entry).reduce(\n\t\t\t(sum, message) => sum + estimateTokens(message),\n\t\t\t0,\n\t\t);\n\t\tif (messageTokens === 0) continue;\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 adjacent metadata entries that do not affect context.\n\twhile (cutIndex > startIndex) {\n\t\tconst prevEntry = entries[cutIndex - 1];\n\t\t// Stop at compaction boundaries or context-visible entries.\n\t\tif (prevEntry.type === \"compaction\" || sessionEntryToContextMessages(prevEntry).length > 0) {\n\t\t\tbreak;\n\t\t}\n\t\tcutIndex--;\n\t}\n\n\t// Determine if this is a split turn\n\tconst cutEntry = entries[cutIndex];\n\tconst startsTurn = isTurnStartEntry(cutEntry);\n\tconst turnStartIndex = startsTurn ? -1 : findTurnStartIndex(entries, cutIndex, startIndex);\n\n\treturn {\n\t\tfirstKeptEntryIndex: cutIndex,\n\t\tturnStartIndex,\n\t\tisSplitTurn: !startsTurn && turnStartIndex !== -1,\n\t};\n}\n\n// ============================================================================\n// Summarization\n// ============================================================================\n\nconst SUMMARIZATION_PROMPT = `The messages above are a conversation to summarize. Create a structured context checkpoint summary that another LLM will use to continue the work.\n\nUse this EXACT format:\n\n## Goal\n[What is the user trying to accomplish? Can be multiple items if the session covers different tasks.]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned by user]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Current work]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [Ordered list of what should happen next]\n\n## Critical Context\n- [Any data, examples, or references needed to continue]\n- [Or \"(none)\" if not applicable]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\nconst UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.\n\nUpdate the existing structured summary with new information. RULES:\n- PRESERVE all existing information from the previous summary\n- ADD new progress, decisions, and context from the new messages\n- UPDATE the Progress section: move items from \"In Progress\" to \"Done\" when completed\n- UPDATE \"Next Steps\" based on what was accomplished\n- PRESERVE exact file paths, function names, and error messages\n- If something is no longer relevant, you may remove it\n\nUse this EXACT format:\n\n## Goal\n[Preserve existing goals, add new ones if the task expanded]\n\n## Constraints & Preferences\n- [Preserve existing, add new ones discovered]\n\n## Progress\n### Done\n- [x] [Include previously done items AND newly completed items]\n\n### In Progress\n- [ ] [Current work - update based on progress]\n\n### Blocked\n- [Current blockers - remove if resolved]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale] (preserve all previous, add new)\n\n## Next Steps\n1. [Update based on current state]\n\n## Critical Context\n- [Preserve important context, add new if needed]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\nfunction createSummarizationOptions(\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tenv: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n): SimpleStreamOptions {\n\tconst options: SimpleStreamOptions = { maxTokens, signal, apiKey, headers, env };\n\tif (model.reasoning && thinkingLevel && thinkingLevel !== \"off\") {\n\t\toptions.reasoning = thinkingLevel;\n\t}\n\treturn options;\n}\n\n/**\n * Shared choke point for every compaction/branch-summary summarization call. Wraps the\n * single LLM call in {@link retryAssistantCall} so transient stream drops (e.g.\n * `terminated`, socket close) honor the configured retry policy instead of failing\n * the whole compaction on the first attempt. Deterministic errors and aborts return\n * immediately (see {@link retryAssistantCall}).\n */\nexport async function completeSummarization(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n): Promise<AssistantMessage> {\n\t// Summaries are standalone requests, so isolate routing and avoid cache writes that cannot be reused.\n\tconst requestOptions: SimpleStreamOptions = {\n\t\t...options,\n\t\tcacheRetention: \"none\",\n\t\tsessionId: uuidv7(),\n\t};\n\tconst produce = async (): Promise<AssistantMessage> =>\n\t\tstreamFn\n\t\t\t? (await streamFn(model, context, requestOptions)).result()\n\t\t\t: completeSimple(model, context, requestOptions);\n\treturn retryAssistantCall(produce, retry, requestOptions.signal, callbacks);\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n): Promise<string> {\n\treturn (\n\t\tawait generateSummaryWithUsage(\n\t\t\tcurrentMessages,\n\t\t\tmodel,\n\t\t\treserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tcustomInstructions,\n\t\t\tpreviousSummary,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tenv,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t)\n\t).text;\n}\n\n/** Generate or update a conversation summary and return its provider usage. */\nexport async function generateSummaryWithUsage(\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n): Promise<{ text: string; usage: Usage }> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.8 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t);\n\n\t// Use update prompt if we have a previous summary, otherwise initial prompt\n\tlet basePrompt = previousSummary ? UPDATE_SUMMARIZATION_PROMPT : SUMMARIZATION_PROMPT;\n\tif (customInstructions) {\n\t\tbasePrompt = `${basePrompt}\\n\\nAdditional focus: ${customInstructions}`;\n\t}\n\n\t// Serialize conversation to text so model doesn't try to continue it\n\t// Convert to LLM messages first (handles custom types like bashExecution, custom, etc.)\n\tconst llmMessages = convertToLlm(currentMessages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build the prompt with conversation wrapped in tags\n\tlet promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n`;\n\tif (previousSummary) {\n\t\tpromptText += `<previous-summary>\\n${previousSummary}\\n</previous-summary>\\n\\n`;\n\t}\n\tpromptText += basePrompt;\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\tconst completionOptions = createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel);\n\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\t{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },\n\t\tcompletionOptions,\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\n\tconst textContent = contentText(response.content);\n\n\treturn { text: textContent, usage: response.usage };\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/** 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\tif (messagesToSummarize.length === 0 && turnPrefixMessages.length === 0) {\n\t\treturn undefined;\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\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\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\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} = preparation;\n\n\t// Generate summaries and merge into one\n\tlet summary: string;\n\tlet summaryUsage: Usage;\n\n\tif (isSplitTurn && turnPrefixMessages.length > 0) {\n\t\tlet historyText = \"No prior history.\";\n\t\tlet historyUsage: Usage | undefined;\n\t\tif (messagesToSummarize.length > 0) {\n\t\t\tconst historyResult = await generateSummaryWithUsage(\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\tcustomInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tenv,\n\t\t\t\tretry,\n\t\t\t\tcallbacks,\n\t\t\t);\n\t\t\thistoryText = historyResult.text;\n\t\t\thistoryUsage = historyResult.usage;\n\t\t}\n\t\tconst turnPrefixResult = 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\tenv,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t);\n\t\t// Merge into single summary\n\t\tsummary = `${historyText}\\n\\n---\\n\\n**Turn Context (split turn):**\\n\\n${turnPrefixResult.text}`;\n\t\tsummaryUsage = historyUsage ? combineUsage(historyUsage, turnPrefixResult.usage) : turnPrefixResult.usage;\n\t} else {\n\t\t// Just generate history summary\n\t\tconst result = await generateSummaryWithUsage(\n\t\t\tmessagesToSummarize,\n\t\t\tmodel,\n\t\t\tsettings.reserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tcustomInstructions,\n\t\t\tpreviousSummary,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tenv,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t);\n\t\tsummary = result.text;\n\t\tsummaryUsage = result.usage;\n\t}\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tusage: summaryUsage,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t};\n}\n\n/**\n * Generate a summary for a turn prefix (when splitting a turn).\n */\nasync function generateTurnPrefixSummary(\n\tmessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tenv?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n): Promise<{ text: string; usage: Usage }> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.5 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t); // Smaller budget for turn prefix\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\tconst promptText = `<conversation>\\n${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, env, signal, thinkingLevel),\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\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 {\n\t\ttext: contentText(response.content),\n\t\tusage: response.usage,\n\t};\n}\n"]}
+{"version":3,"file":"compaction.d.ts","sourceRoot":"","sources":["../../../src/core/compaction/compaction.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAe,KAAK,cAAc,EAAE,KAAK,WAAW,EAA8B,MAAM,uBAAuB,CAAC;AACvH,OAAO,KAAK,EAAE,gBAAgB,EAAE,OAAO,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,EAAE,MAAM,8BAA8B,CAAC;AAEjH,OAAO,KAAK,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAElF,OAAO,EAGN,KAAK,YAAY,EAEjB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAIN,KAAK,cAAc,EAInB,MAAM,YAAY,CAAC;AAMpB,kEAAkE;AAClE,MAAM,WAAW,iBAAiB;IACjC,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACxB;AAkDD,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,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B,2EAA2E;IAC3E,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,+FAA+F;IAC/F,OAAO,CAAC,EAAE,CAAC,CAAC;CACZ;AA6BD,MAAM,WAAW,kBAAkB;IAClC,OAAO,EAAE,OAAO,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,gBAAgB,EAAE,MAAM,CAAC;CACzB;AAED,eAAO,MAAM,2BAA2B,EAAE,kBAIzC,CAAC;AAMF;;;GAGG;AACH,wBAAgB,sBAAsB,CAAC,KAAK,EAAE,KAAK,GAAG,MAAM,CAE3D;AAqBD;;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;;GAEG;AACH,wBAAgB,aAAa,CAAC,aAAa,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,EAAE,QAAQ,EAAE,kBAAkB,GAAG,OAAO,CAGjH;AAwBD;;;GAGG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,YAAY,GAAG,MAAM,CAwC5D;AA2DD;;;GAGG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,YAAY,EAAE,EAAE,UAAU,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,MAAM,CAO1G;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,CAqDhB;AAiGD;;;;;;GAMG;AACH,wBAAsB,qBAAqB,CAC1C,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,EACjB,OAAO,EAAE,OAAO,EAChB,OAAO,EAAE,mBAAmB,EAC5B,QAAQ,CAAC,EAAE,QAAQ,EACnB,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,GACxB,OAAO,CAAC,gBAAgB,CAAC,CAc3B;AAED;;;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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,EAC1B,SAAS,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC,MAAM,CAAC,CAmBjB;AAgBD,+EAA+E;AAC/E,wBAAsB,wBAAwB,CAC7C,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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,EAC1B,SAAS,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,KAAK,CAAA;CAAE,CAAC,CAsDzC;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,8CAA8C;IAC9C,QAAQ,EAAE,kBAAkB,CAAC;CAC7B;AAED,wBAAgB,iBAAiB,CAChC,WAAW,EAAE,YAAY,EAAE,EAC3B,QAAQ,EAAE,kBAAkB,GAC1B,qBAAqB,GAAG,SAAS,CA4EnC;AAqBD;;;;;;;GAOG;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,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,EAC5B,KAAK,CAAC,EAAE,WAAW,EACnB,SAAS,CAAC,EAAE,cAAc,EAC1B,SAAS,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC,gBAAgB,CAAC,CA6F3B","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 { contentText, type RetryCallbacks, type RetryPolicy, retryAssistantCall, uuidv7 } from \"@earendil-works/pi-ai\";\nimport type { AssistantMessage, Context, Model, SimpleStreamOptions, Usage } from \"@earendil-works/pi-ai/compat\";\nimport { completeSimple } from \"@earendil-works/pi-ai/compat\";\nimport type { AgentMessage, StreamFn, ThinkingLevel } from \"apex-code-agent-core\";\nimport { convertToLlm } from \"../messages.ts\";\nimport {\n\tbuildSessionContext,\n\ttype CompactionEntry,\n\ttype SessionEntry,\n\tsessionEntryToContextMessages,\n} from \"../session-manager.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// 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 getMessageFromEntryForCompaction(entry: SessionEntry): AgentMessage | undefined {\n\tif (entry.type === \"compaction\") {\n\t\treturn undefined;\n\t}\n\treturn sessionEntryToContextMessages(entry)[0];\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\testimatedTokensAfter?: number;\n\t/** Usage from the LLM call(s) that generated this summary, if available */\n\tusage?: Usage;\n\t/** Extension-specific data (e.g., ArtifactIndex, version markers for structured compaction) */\n\tdetails?: T;\n}\n\nfunction combineUsage(first: Usage, second: Usage): Usage {\n\treturn {\n\t\tinput: first.input + second.input,\n\t\toutput: first.output + second.output,\n\t\tcacheRead: first.cacheRead + second.cacheRead,\n\t\tcacheWrite: first.cacheWrite + second.cacheWrite,\n\t\t...(first.cacheWrite1h !== undefined || second.cacheWrite1h !== undefined\n\t\t\t? { cacheWrite1h: (first.cacheWrite1h ?? 0) + (second.cacheWrite1h ?? 0) }\n\t\t\t: {}),\n\t\t...(first.reasoning !== undefined || second.reasoning !== undefined\n\t\t\t? { reasoning: (first.reasoning ?? 0) + (second.reasoning ?? 0) }\n\t\t\t: {}),\n\t\ttotalTokens: first.totalTokens + second.totalTokens,\n\t\tcost: {\n\t\t\tinput: first.cost.input + second.cost.input,\n\t\t\toutput: first.cost.output + second.cost.output,\n\t\t\tcacheRead: first.cost.cacheRead + second.cost.cacheRead,\n\t\t\tcacheWrite: first.cost.cacheWrite + second.cost.cacheWrite,\n\t\t\ttotal: first.cost.total + second.cost.total,\n\t\t},\n\t};\n}\n\n// ============================================================================\n// Types\n// ============================================================================\n\nexport interface CompactionSettings {\n\tenabled: boolean;\n\treserveTokens: number;\n\tkeepRecentTokens: number;\n}\n\nexport const DEFAULT_COMPACTION_SETTINGS: CompactionSettings = {\n\tenabled: true,\n\treserveTokens: 16384,\n\tkeepRecentTokens: 20000,\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, error, and all-zero usage 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 (\n\t\t\tassistantMsg.stopReason !== \"aborted\" &&\n\t\t\tassistantMsg.stopReason !== \"error\" &&\n\t\t\tassistantMsg.usage &&\n\t\t\tcalculateContextTokens(assistantMsg.usage) > 0\n\t\t) {\n\t\t\treturn assistantMsg.usage;\n\t\t}\n\t}\n\treturn undefined;\n}\n\n/**\n * Find the last valid 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 * Check if compaction should trigger based on context usage.\n */\nexport function shouldCompact(contextTokens: number, contextWindow: number, settings: CompactionSettings): boolean {\n\tif (!settings.enabled) return false;\n\treturn contextTokens > contextWindow - settings.reserveTokens;\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\nfunction isCutPointMessage(message: AgentMessage): boolean {\n\tswitch (message.role) {\n\t\tcase \"user\":\n\t\tcase \"assistant\":\n\t\tcase \"bashExecution\":\n\t\tcase \"custom\":\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\":\n\t\t\treturn true;\n\t\tcase \"toolResult\":\n\t\t\treturn false;\n\t}\n\treturn false;\n}\n\nfunction isTurnStartMessage(message: AgentMessage): boolean {\n\tswitch (message.role) {\n\t\tcase \"user\":\n\t\tcase \"bashExecution\":\n\t\tcase \"custom\":\n\t\tcase \"branchSummary\":\n\t\tcase \"compactionSummary\":\n\t\t\treturn true;\n\t\tcase \"assistant\":\n\t\tcase \"toolResult\":\n\t\t\treturn false;\n\t}\n\treturn false;\n}\n\nfunction isTurnStartEntry(entry: SessionEntry): boolean {\n\tif (entry.type === \"compaction\") {\n\t\treturn false;\n\t}\n\treturn sessionEntryToContextMessages(entry).some(isTurnStartMessage);\n}\n\n/**\n * Find valid cut points: indices of context-visible user-like or assistant 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 */\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\tif (entry.type === \"compaction\") {\n\t\t\tcontinue;\n\t\t}\n\t\tif (sessionEntryToContextMessages(entry).some(isCutPointMessage)) {\n\t\t\tcutPoints.push(i);\n\t\t}\n\t}\n\treturn cutPoints;\n}\n\n/**\n * Find the context-visible user-role message that starts the turn containing the given entry index.\n * Returns -1 if no turn start found before the index.\n */\nexport function findTurnStartIndex(entries: SessionEntry[], entryIndex: number, startIndex: number): number {\n\tfor (let i = entryIndex; i >= startIndex; i--) {\n\t\tif (isTurnStartEntry(entries[i])) {\n\t\t\treturn i;\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\tconst messageTokens = sessionEntryToContextMessages(entry).reduce(\n\t\t\t(sum, message) => sum + estimateTokens(message),\n\t\t\t0,\n\t\t);\n\t\tif (messageTokens === 0) continue;\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 adjacent metadata entries that do not affect context.\n\twhile (cutIndex > startIndex) {\n\t\tconst prevEntry = entries[cutIndex - 1];\n\t\t// Stop at compaction boundaries or context-visible entries.\n\t\tif (prevEntry.type === \"compaction\" || sessionEntryToContextMessages(prevEntry).length > 0) {\n\t\t\tbreak;\n\t\t}\n\t\tcutIndex--;\n\t}\n\n\t// Determine if this is a split turn\n\tconst cutEntry = entries[cutIndex];\n\tconst startsTurn = isTurnStartEntry(cutEntry);\n\tconst turnStartIndex = startsTurn ? -1 : findTurnStartIndex(entries, cutIndex, startIndex);\n\n\treturn {\n\t\tfirstKeptEntryIndex: cutIndex,\n\t\tturnStartIndex,\n\t\tisSplitTurn: !startsTurn && turnStartIndex !== -1,\n\t};\n}\n\n// ============================================================================\n// Summarization\n// ============================================================================\n\nconst SUMMARIZATION_PROMPT = `The messages above are a conversation to summarize. Create a structured context checkpoint summary that another LLM will use to continue the work.\n\nUse this EXACT format:\n\n## Goal\n[What is the user trying to accomplish? Can be multiple items if the session covers different tasks.]\n\n## Constraints & Preferences\n- [Any constraints, preferences, or requirements mentioned by user]\n- [Or \"(none)\" if none were mentioned]\n\n## Progress\n### Done\n- [x] [Completed tasks/changes]\n\n### In Progress\n- [ ] [Current work]\n\n### Blocked\n- [Issues preventing progress, if any]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale]\n\n## Next Steps\n1. [Ordered list of what should happen next]\n\n## Critical Context\n- [Any data, examples, or references needed to continue]\n- [Or \"(none)\" if not applicable]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\nconst UPDATE_SUMMARIZATION_INSTRUCTIONS = `Update the existing structured summary with new information. RULES:\n- PRESERVE all existing information from the previous summary\n- ADD new progress, decisions, and context from the new messages\n- UPDATE the Progress section: move items from \"In Progress\" to \"Done\" when completed\n- UPDATE \"Next Steps\" based on what was accomplished\n- PRESERVE exact file paths, function names, and error messages\n- If something is no longer relevant, you may remove it\n\nUse this EXACT format:\n\n## Goal\n[Preserve existing goals, add new ones if the task expanded]\n\n## Constraints & Preferences\n- [Preserve existing, add new ones discovered]\n\n## Progress\n### Done\n- [x] [Include previously done items AND newly completed items]\n\n### In Progress\n- [ ] [Current work - update based on progress]\n\n### Blocked\n- [Current blockers - remove if resolved]\n\n## Key Decisions\n- **[Decision]**: [Brief rationale] (preserve all previous, add new)\n\n## Next Steps\n1. [Update based on current state]\n\n## Critical Context\n- [Preserve important context, add new if needed]\n\nKeep each section concise. Preserve exact file paths, function names, and error messages.`;\n\nconst UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.\n\n${UPDATE_SUMMARIZATION_INSTRUCTIONS}`;\n\nfunction createSummarizationOptions(\n\tmodel: Model<any>,\n\tmaxTokens: number,\n\tapiKey: string | undefined,\n\theaders: Record<string, string> | undefined,\n\tenv: Record<string, string> | undefined,\n\tsignal: AbortSignal | undefined,\n\tthinkingLevel: ThinkingLevel | undefined,\n\tsessionId: string | undefined,\n): SimpleStreamOptions {\n\tconst options: SimpleStreamOptions = { maxTokens, signal, apiKey, headers, env, sessionId };\n\tif (model.reasoning && thinkingLevel && thinkingLevel !== \"off\") {\n\t\toptions.reasoning = thinkingLevel;\n\t}\n\treturn options;\n}\n\n/**\n * Shared choke point for every compaction/branch-summary summarization call. Wraps the\n * single LLM call in {@link retryAssistantCall} so transient stream drops (e.g.\n * `terminated`, socket close) honor the configured retry policy instead of failing\n * the whole compaction on the first attempt. Deterministic errors and aborts return\n * immediately (see {@link retryAssistantCall}).\n */\nexport async function completeSummarization(\n\tmodel: Model<any>,\n\tcontext: Context,\n\toptions: SimpleStreamOptions,\n\tstreamFn?: StreamFn,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n): Promise<AssistantMessage> {\n\t// Avoid cache writes for one-off summaries. Reuse caller-supplied routing when available;\n\t// callers without a session ID, including branch summaries, receive a fresh routing ID.\n\tconst requestOptions: SimpleStreamOptions = {\n\t\t...options,\n\t\tcacheRetention: \"none\",\n\t\tsessionId: options.sessionId ?? uuidv7(),\n\t\ttoolChoice: \"none\",\n\t};\n\tconst produce = async (): Promise<AssistantMessage> =>\n\t\tstreamFn\n\t\t\t? (await streamFn(model, context, requestOptions)).result()\n\t\t\t: completeSimple(model, context, requestOptions);\n\treturn retryAssistantCall(produce, retry, requestOptions.signal, callbacks);\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n\tsessionId?: string,\n): Promise<string> {\n\treturn (\n\t\tawait generateSummaryWithUsage(\n\t\t\tcurrentMessages,\n\t\t\tmodel,\n\t\t\treserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tcustomInstructions,\n\t\t\tpreviousSummary,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tenv,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t\tsessionId,\n\t\t)\n\t).text;\n}\n\n/** Build the provider context for a standalone summary request. */\nfunction buildSummarizationContext(promptText: string): Context {\n\treturn {\n\t\tsystemPrompt: SUMMARIZATION_SYSTEM_PROMPT,\n\t\tmessages: [\n\t\t\t{\n\t\t\t\trole: \"user\",\n\t\t\t\tcontent: [{ type: \"text\", text: promptText }],\n\t\t\t\ttimestamp: Date.now(),\n\t\t\t},\n\t\t],\n\t};\n}\n\n/** Generate or update a conversation summary and return its provider usage. */\nexport async function generateSummaryWithUsage(\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n\tsessionId?: string,\n): Promise<{ text: string; usage: Usage }> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.8 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t);\n\n\t// Use update prompt if we have a previous summary, otherwise initial prompt\n\tlet basePrompt = previousSummary ? UPDATE_SUMMARIZATION_PROMPT : SUMMARIZATION_PROMPT;\n\tif (customInstructions) {\n\t\tbasePrompt = `${basePrompt}\\n\\nAdditional focus: ${customInstructions}`;\n\t}\n\n\t// Serialize conversation to text so model doesn't try to continue it\n\t// Convert to LLM messages first (handles custom types like bashExecution, custom, etc.)\n\tconst llmMessages = convertToLlm(currentMessages);\n\tconst conversationText = serializeConversation(llmMessages);\n\n\t// Build the prompt with conversation wrapped in tags\n\tlet promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n`;\n\tif (previousSummary) {\n\t\tpromptText += `<previous-summary>\\n${previousSummary}\\n</previous-summary>\\n\\n`;\n\t}\n\tpromptText += basePrompt;\n\n\tconst completionOptions = createSummarizationOptions(\n\t\tmodel,\n\t\tmaxTokens,\n\t\tapiKey,\n\t\theaders,\n\t\tenv,\n\t\tsignal,\n\t\tthinkingLevel,\n\t\tsessionId,\n\t);\n\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\tbuildSummarizationContext(promptText),\n\t\tcompletionOptions,\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\tif (response.content.some((block) => block.type === \"toolCall\")) {\n\t\tthrow new Error(\"Summarization attempted to call a tool\");\n\t}\n\n\tconst textContent = contentText(response.content);\n\n\treturn { text: textContent, usage: response.usage };\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/** 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\tif (messagesToSummarize.length === 0 && turnPrefixMessages.length === 0) {\n\t\treturn undefined;\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\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\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 * @param sessionId - Optional routing session ID forwarded without enabling prompt caching\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\tenv?: Record<string, string>,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n\tsessionId?: string,\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} = preparation;\n\n\t// Generate summaries and merge into one\n\tlet summary: string;\n\tlet summaryUsage: Usage;\n\n\tif (isSplitTurn && turnPrefixMessages.length > 0) {\n\t\tlet historyText = \"No prior history.\";\n\t\tlet historyUsage: Usage | undefined;\n\t\tif (messagesToSummarize.length > 0) {\n\t\t\tconst historyResult = await generateSummaryWithUsage(\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\tcustomInstructions,\n\t\t\t\tpreviousSummary,\n\t\t\t\tthinkingLevel,\n\t\t\t\tstreamFn,\n\t\t\t\tenv,\n\t\t\t\tretry,\n\t\t\t\tcallbacks,\n\t\t\t\tsessionId,\n\t\t\t);\n\t\t\thistoryText = historyResult.text;\n\t\t\thistoryUsage = historyResult.usage;\n\t\t}\n\t\tconst turnPrefixResult = 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\tenv,\n\t\t\tsignal,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t\tsessionId,\n\t\t);\n\t\t// Merge into single summary\n\t\tsummary = `${historyText}\\n\\n---\\n\\n**Turn Context (split turn):**\\n\\n${turnPrefixResult.text}`;\n\t\tsummaryUsage = historyUsage ? combineUsage(historyUsage, turnPrefixResult.usage) : turnPrefixResult.usage;\n\t} else {\n\t\t// Just generate history summary\n\t\tconst result = await generateSummaryWithUsage(\n\t\t\tmessagesToSummarize,\n\t\t\tmodel,\n\t\t\tsettings.reserveTokens,\n\t\t\tapiKey,\n\t\t\theaders,\n\t\t\tsignal,\n\t\t\tcustomInstructions,\n\t\t\tpreviousSummary,\n\t\t\tthinkingLevel,\n\t\t\tstreamFn,\n\t\t\tenv,\n\t\t\tretry,\n\t\t\tcallbacks,\n\t\t\tsessionId,\n\t\t);\n\t\tsummary = result.text;\n\t\tsummaryUsage = result.usage;\n\t}\n\n\t// Compute file lists and append to summary\n\tconst { readFiles, modifiedFiles } = computeFileLists(fileOps);\n\tsummary += formatFileOperations(readFiles, modifiedFiles);\n\n\tif (!firstKeptEntryId) {\n\t\tthrow new Error(\"First kept entry has no UUID - session may need migration\");\n\t}\n\n\treturn {\n\t\tsummary,\n\t\tfirstKeptEntryId,\n\t\ttokensBefore,\n\t\tusage: summaryUsage,\n\t\tdetails: { readFiles, modifiedFiles } as CompactionDetails,\n\t};\n}\n\n/**\n * Generate a summary for a turn prefix (when splitting a turn).\n */\nasync function generateTurnPrefixSummary(\n\tmessages: AgentMessage[],\n\tmodel: Model<any>,\n\treserveTokens: number,\n\tapiKey: string | undefined,\n\theaders?: Record<string, string>,\n\tenv?: Record<string, string>,\n\tsignal?: AbortSignal,\n\tthinkingLevel?: ThinkingLevel,\n\tstreamFn?: StreamFn,\n\tretry?: RetryPolicy,\n\tcallbacks?: RetryCallbacks,\n\tsessionId?: string,\n): Promise<{ text: string; usage: Usage }> {\n\tconst maxTokens = Math.min(\n\t\tMath.floor(0.5 * reserveTokens),\n\t\tmodel.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY,\n\t); // Smaller budget for turn prefix\n\tconst llmMessages = convertToLlm(messages);\n\tconst conversationText = serializeConversation(llmMessages);\n\tconst promptText = `<conversation>\\n${conversationText}\\n</conversation>\\n\\n${TURN_PREFIX_SUMMARIZATION_PROMPT}`;\n\n\tconst response = await completeSummarization(\n\t\tmodel,\n\t\tbuildSummarizationContext(promptText),\n\t\tcreateSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel, sessionId),\n\t\tstreamFn,\n\t\tretry,\n\t\tcallbacks,\n\t);\n\n\tif (response.stopReason === \"error\") {\n\t\tthrow new Error(`Turn prefix summarization failed: ${response.errorMessage || \"Unknown error\"}`);\n\t}\n\tif (response.content.some((block) => block.type === \"toolCall\")) {\n\t\tthrow new Error(\"Turn prefix summarization attempted to call a tool\");\n\t}\n\n\treturn {\n\t\ttext: contentText(response.content),\n\t\tusage: response.usage,\n\t};\n}\n"]}

package/dist/core/compaction/compaction.js

@@ -385,9 +385,7 @@ Use this EXACT format:
 - [Or "(none)" if not applicable]
 
 Keep each section concise. Preserve exact file paths, function names, and error messages.`;
-const UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.
-
-Update the existing structured summary with new information. RULES:
+const UPDATE_SUMMARIZATION_INSTRUCTIONS = `Update the existing structured summary with new information. RULES:
 - PRESERVE all existing information from the previous summary
 - ADD new progress, decisions, and context from the new messages
 - UPDATE the Progress section: move items from "In Progress" to "Done" when completed
@@ -423,8 +421,11 @@ Use this EXACT format:
 - [Preserve important context, add new if needed]
 
 Keep each section concise. Preserve exact file paths, function names, and error messages.`;
-function createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel) {
-    const options = { maxTokens, signal, apiKey, headers, env };
+const UPDATE_SUMMARIZATION_PROMPT = `The messages above are NEW conversation messages to incorporate into the existing summary provided in <previous-summary> tags.
+
+${UPDATE_SUMMARIZATION_INSTRUCTIONS}`;
+function createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel, sessionId) {
+    const options = { maxTokens, signal, apiKey, headers, env, sessionId };
     if (model.reasoning && thinkingLevel && thinkingLevel !== "off") {
         options.reasoning = thinkingLevel;
     }
@@ -438,11 +439,13 @@ function createSummarizationOptions(model, maxTokens, apiKey, headers, env, sign
  * immediately (see {@link retryAssistantCall}).
  */
 export async function completeSummarization(model, context, options, streamFn, retry, callbacks) {
-    // Summaries are standalone requests, so isolate routing and avoid cache writes that cannot be reused.
+    // Avoid cache writes for one-off summaries. Reuse caller-supplied routing when available;
+    // callers without a session ID, including branch summaries, receive a fresh routing ID.
     const requestOptions = {
         ...options,
         cacheRetention: "none",
-        sessionId: uuidv7(),
+        sessionId: options.sessionId ?? uuidv7(),
+        toolChoice: "none",
     };
     const produce = async () => streamFn
         ? (await streamFn(model, context, requestOptions)).result()
@@ -453,11 +456,24 @@ export async function completeSummarization(model, context, options, streamFn, r
  * Generate a summary of the conversation using the LLM.
  * If previousSummary is provided, uses the update prompt to merge.
  */
-export async function generateSummary(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks) {
-    return (await generateSummaryWithUsage(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks)).text;
+export async function generateSummary(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks, sessionId) {
+    return (await generateSummaryWithUsage(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks, sessionId)).text;
+}
+/** Build the provider context for a standalone summary request. */
+function buildSummarizationContext(promptText) {
+    return {
+        systemPrompt: SUMMARIZATION_SYSTEM_PROMPT,
+        messages: [
+            {
+                role: "user",
+                content: [{ type: "text", text: promptText }],
+                timestamp: Date.now(),
+            },
+        ],
+    };
 }
 /** Generate or update a conversation summary and return its provider usage. */
-export async function generateSummaryWithUsage(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks) {
+export async function generateSummaryWithUsage(currentMessages, model, reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks, sessionId) {
     const maxTokens = Math.min(Math.floor(0.8 * reserveTokens), model.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY);
     // Use update prompt if we have a previous summary, otherwise initial prompt
     let basePrompt = previousSummary ? UPDATE_SUMMARIZATION_PROMPT : SUMMARIZATION_PROMPT;
@@ -474,18 +490,14 @@ export async function generateSummaryWithUsage(currentMessages, model, reserveTo
         promptText += `<previous-summary>\n${previousSummary}\n</previous-summary>\n\n`;
     }
     promptText += basePrompt;
-    const summarizationMessages = [
-        {
-            role: "user",
-            content: [{ type: "text", text: promptText }],
-            timestamp: Date.now(),
-        },
-    ];
-    const completionOptions = createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel);
-    const response = await completeSummarization(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, completionOptions, streamFn, retry, callbacks);
+    const completionOptions = createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel, sessionId);
+    const response = await completeSummarization(model, buildSummarizationContext(promptText), completionOptions, streamFn, retry, callbacks);
     if (response.stopReason === "error") {
         throw new Error(`Summarization failed: ${response.errorMessage || "Unknown error"}`);
     }
+    if (response.content.some((block) => block.type === "toolCall")) {
+        throw new Error("Summarization attempted to call a tool");
+    }
     const textContent = contentText(response.content);
     return { text: textContent, usage: response.usage };
 }
@@ -579,8 +591,9 @@ Be concise. Focus on what's needed to understand the kept suffix.`;
  *
  * @param preparation - Pre-calculated preparation from prepareCompaction()
  * @param customInstructions - Optional custom focus for the summary
+ * @param sessionId - Optional routing session ID forwarded without enabling prompt caching
  */
-export async function compact(preparation, model, apiKey, headers, customInstructions, signal, thinkingLevel, streamFn, env, retry, callbacks) {
+export async function compact(preparation, model, apiKey, headers, customInstructions, signal, thinkingLevel, streamFn, env, retry, callbacks, sessionId) {
     const { firstKeptEntryId, messagesToSummarize, turnPrefixMessages, isSplitTurn, tokensBefore, previousSummary, fileOps, settings, } = preparation;
     // Generate summaries and merge into one
     let summary;
@@ -589,18 +602,18 @@ export async function compact(preparation, model, apiKey, headers, customInstruc
         let historyText = "No prior history.";
         let historyUsage;
         if (messagesToSummarize.length > 0) {
-            const historyResult = await generateSummaryWithUsage(messagesToSummarize, model, settings.reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks);
+            const historyResult = await generateSummaryWithUsage(messagesToSummarize, model, settings.reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks, sessionId);
             historyText = historyResult.text;
             historyUsage = historyResult.usage;
         }
-        const turnPrefixResult = await generateTurnPrefixSummary(turnPrefixMessages, model, settings.reserveTokens, apiKey, headers, env, signal, thinkingLevel, streamFn, retry, callbacks);
+        const turnPrefixResult = await generateTurnPrefixSummary(turnPrefixMessages, model, settings.reserveTokens, apiKey, headers, env, signal, thinkingLevel, streamFn, retry, callbacks, sessionId);
         // Merge into single summary
         summary = `${historyText}\n\n---\n\n**Turn Context (split turn):**\n\n${turnPrefixResult.text}`;
         summaryUsage = historyUsage ? combineUsage(historyUsage, turnPrefixResult.usage) : turnPrefixResult.usage;
     }
     else {
         // Just generate history summary
-        const result = await generateSummaryWithUsage(messagesToSummarize, model, settings.reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks);
+        const result = await generateSummaryWithUsage(messagesToSummarize, model, settings.reserveTokens, apiKey, headers, signal, customInstructions, previousSummary, thinkingLevel, streamFn, env, retry, callbacks, sessionId);
         summary = result.text;
         summaryUsage = result.usage;
     }
@@ -621,22 +634,18 @@ export async function compact(preparation, model, apiKey, headers, customInstruc
 /**
  * Generate a summary for a turn prefix (when splitting a turn).
  */
-async function generateTurnPrefixSummary(messages, model, reserveTokens, apiKey, headers, env, signal, thinkingLevel, streamFn, retry, callbacks) {
+async function generateTurnPrefixSummary(messages, model, reserveTokens, apiKey, headers, env, signal, thinkingLevel, streamFn, retry, callbacks, sessionId) {
     const maxTokens = Math.min(Math.floor(0.5 * reserveTokens), model.maxTokens > 0 ? model.maxTokens : Number.POSITIVE_INFINITY); // Smaller budget for turn prefix
     const llmMessages = convertToLlm(messages);
     const conversationText = serializeConversation(llmMessages);
     const promptText = `<conversation>\n${conversationText}\n</conversation>\n\n${TURN_PREFIX_SUMMARIZATION_PROMPT}`;
-    const summarizationMessages = [
-        {
-            role: "user",
-            content: [{ type: "text", text: promptText }],
-            timestamp: Date.now(),
-        },
-    ];
-    const response = await completeSummarization(model, { systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages }, createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel), streamFn, retry, callbacks);
+    const response = await completeSummarization(model, buildSummarizationContext(promptText), createSummarizationOptions(model, maxTokens, apiKey, headers, env, signal, thinkingLevel, sessionId), streamFn, retry, callbacks);
     if (response.stopReason === "error") {
         throw new Error(`Turn prefix summarization failed: ${response.errorMessage || "Unknown error"}`);
     }
+    if (response.content.some((block) => block.type === "toolCall")) {
+        throw new Error("Turn prefix summarization attempted to call a tool");
+    }
     return {
         text: contentText(response.content),
         usage: response.usage,