pi-smart-compact 7.16.0 → 7.18.0

package/CHANGELOG.md

@@ -1,5 +1,47 @@
 # Changelog
 
+## [7.18.0] - 2026-06-26
+
+### Fixed
+- **`mergeExtractions` mainGoal corruption** — incremental extraction replaced the original goal with a delta-suffix message; now prefers `base.mainGoal`.
+- **`mergeExtractions` boundary gap** — `lastUserMessages`/`lastErrors` now span the cache boundary instead of dropping base's tail.
+- **`catalogErrors` id-less retry** — retry resolution falls back to tool-name match when the retry call lacks an id.
+- **Pruning hides errors** — short error-containing tool outputs preserved from 800-char truncation (shared `LIKELY_ERROR_RE`).
+- **`computeDelta` collision** — prefix-slice keys collided different items sharing an opening; now full-normalized text.
+- **`smartKeepBoundary` dead regex** — `path="..."` never matched agent output; replaced with toolCall-arg file extraction.
+- **`buildCompactionState` basename** — raw basename → `buildPathNeedles` (generic-basename gate, consistent with `extractOpenLoops`).
+- **`extractNextActions`/`Critical`** — case-sensitive regex → canonical `findSection`.
+- **`CONSTRAINT_PATTERNS`** — `\b` → lookarounds so Turkish-leading chars (`önemli`, `şart`) match.
+- **`verifySummary` fuzzy** — salient-token keyword extraction; error snippet whitespace-normalized.
+- **`estimateTokens` JSON** — density fallback for non-brace-first JSON (capped at 8KB).
+
+### Added
+- **`domain/keywords.ts`** — `extractCheckKeywords` (shared by verify + damage).
+- **Centralized constants** — `TRUNC` (truncation budgets), `ID_PREFIX`, `TUNING` (EMA/clamp/confidence), shared durations, `EXTRACTION_CACHE_PREFIX`, `LIKELY_ERROR_RE`.
+
+### Changed
+- **Zero inline magic values** — ~70 truncation literals, ID prefixes, confidence scores, TTL duplications (7d×3, 1h×2), and tuning factors centralized into named constants.
+- **`makeCompactSessionId`** — single source (services.ts); cache.ts fallback deduped.
+- **`damage.ts` re-question** — shared `extractCheckKeywords` (threshold relaxed for high-precision salient tokens).
+
+## [7.17.0] - 2026-06-26
+
+### Added
+- **Name-agnostic tool classification** (`src/domain/tool-semantics.ts`) — a pure `classifyTool(args)` that classifies a tool call by its argument shape (`mutates` / `accesses` / `executes` / `other`) plus `extractToolPath(args)`. A tool is a write because its arguments carry a content payload, not because its name contains "write" — so this auto-adapts to tools the code has never seen (`hypa_*`, MCP servers, custom extensions) with no name list to maintain.
+
+### Changed
+- **Extraction is now name-agnostic.** `trackFileOps`, `catalogErrors`, `segmentTopicsHeuristic` (`utils/extraction.ts`), `get_file_changes` (`phases/explore.ts`), and re-read detection (`utils/damage.ts`) all classify via `classifyTool` instead of substring name matching. Removes the `WRITE/DELETE/READ_TOOL_HINTS` lists, `hasToolHint`, and the hardcoded `=== "bash"` gate. Shell-like tools (`hypa_shell`, …) and path-bearing readers (`hypa_grep`/`find`/`ls`) are now detected correctly by argument shape.
+- **Model resolution deduplicated** (`index.ts`) — the `provider/id` split + `modelRegistry.find` pattern (duplicated three times) is now a single `findModelById` helper; `resolveModelArg` removed.
+- **`SHIFT_RE` Turkish coverage** — the topic-shift cue regex was ASCII-only and silently missed natural Turkish spellings (`şimdi`, `geçelim`, `bakalım`, `yapalım`, `başka`); both spellings now match, consistent with `FOLLOWUP_RE`'s dual-spelling convention.
+- **`CONSTRAINT_PATTERNS` readability** — the Turkish regexes use raw UTF-8 characters instead of `\u00f6`/`\u015f` escapes (the source is UTF-8; the escapes added no value and hurt readability).
+
+### Fixed
+- **`trackFileOps` duplicated branch** — the `isTruncated` and `!NO_OP_RE` arms had identical bodies; collapsed into one condition (`isTruncated(resultText) || !NO_OP_RE.test(resultText)`).
+- **Dead locals in `segmentTopicsHeuristic`** — `type` and `primaryFile` were assigned every iteration but never read (the topic push uses `currentType`/`currentPrimaryFile`); removed. Path lookup now uses `extractToolPath`, matching the other consumers.
+
+### Tests
+- New `test/tool-semantics.test.ts` covering all four tool classes, the path-only delete/read ambiguity, and `extractToolPath` across key variants. Extraction fixtures updated to realistic content payloads; a regression case proves an unknown tool name (`totally_unknown_mcp_tool`) is still classified correctly. Suite: 501 tests across 43 files.
+
 ## [7.16.0] - 2026-06-18
 
 ### Added

package/dist/constants.d.ts

@@ -8,7 +8,7 @@ import type { CompressionProfile, ProfileConfig } from "./types.ts";
  * `package.json#version`. Do not hand-edit this line for releases; bump
  * package.json and run `bun run sync-version`.
  */
-export declare const VERSION = "7.16.0";
+export declare const VERSION = "7.18.0";
 export declare const CHARS_PER_TOKEN = 3.8;
 export declare const COMPACT_SYSTEM_PREFIX: string;
 export declare const PROFILES: Record<CompressionProfile, ProfileConfig>;
@@ -50,11 +50,66 @@ export declare const MIN_TOKEN_THRESHOLD = 5000;
 export declare const MAX_EXPLORATION_ROUNDS = 8;
 export declare const BACKUP_MAX_FILES = 20;
 export declare const BACKUP_MAX_AGE_MS: number;
-export declare const SUMMARY_SNIPPET_LEN = 100;
-export declare const SHORT_SUMMARY_LEN = 60;
-export declare const ERROR_FUZZY_MATCH_LEN = 30;
+export declare const FIVE_MINUTES_MS: number;
+export declare const ONE_HOUR_MS: number;
+export declare const SEVEN_DAYS_MS: number;
+export declare const THIRTY_DAYS_MS: number;
+export declare const EXTRACTION_CACHE_PREFIX = "compact-extraction-";
+export declare const ID_PREFIX: {
+    readonly PROJECT: "proj-";
+    readonly COMPACT_SESSION: "sc-";
+    readonly MULTI_TOOL_USE_SYNTHETIC: "mtu_";
+    readonly OPEN_LOOP: "loop-";
+    readonly DECISION: "decision-";
+    readonly ERROR: "error-";
+};
+export declare const TUNING: {
+    readonly EMA_PREV: 0.7;
+    readonly EMA_SAMPLE: 0.3;
+    readonly CALIBRATION_CLAMP_MIN: 0.3;
+    readonly CALIBRATION_CLAMP_MAX: 3;
+    readonly CONFIDENCE_HIGH: 0.85;
+    readonly CONFIDENCE_MEDIUM: 0.8;
+    readonly CONFIDENCE_LOW: 0.6;
+};
+export declare const TRUNC: {
+    readonly MESSAGE: 300;
+    readonly ERROR_DETAIL: 500;
+    readonly DECISION_SUMMARY: 200;
+    readonly USER_RESPONSE: 300;
+    readonly CONSTRAINT_TEXT: 300;
+    readonly OPEN_LOOP_SUMMARY: 120;
+    readonly TIMELINE_EVENT: 150;
+    readonly TIMELINE_ERROR: 100;
+    readonly SNIPPET: 80;
+    readonly PREVIEW: 150;
+    readonly PREVIEW_MID: 200;
+    readonly DETAIL: 300;
+    readonly PREVIEW_LONG: 400;
+    readonly PREVIEW_XL: 500;
+    readonly CHUNK_FALLBACK: 180;
+    readonly DECISION_DETAIL: 60;
+    readonly TOPIC_LABEL: 100;
+    readonly PROJ_ID_HASH: 12;
+    readonly CONV_HASH: 8;
+    readonly RESULT_GAPS: 5;
+    readonly SESSION_ID_DISPLAY: 20;
+    readonly ERROR_SNIPPET: 30;
+    readonly TIMELINE_DISPLAY: 10;
+    readonly EXPLORE_RESULTS: 15;
+    readonly BACKUP_PREVIEW_LINES: 5;
+    readonly FINGERPRINT_SEG: 2;
+};
 export declare const DAMAGE_RECENT_MSG_WINDOW = 15;
 export declare const MAX_TOOL_OUTPUT_CHARS = 800;
+/** Shell-output patterns signalling a likely error even in a non-`isError` result. */
+export declare const LIKELY_ERROR_RE: RegExp;
+/** catalogErrors only scans outputs shorter than this — pruning must not truncate below it. */
+export declare const ERROR_SCAN_MAX_LEN = 2000;
+/** How far forward to look for a retry of the same tool after an error. */
+export declare const ERROR_RETRY_WINDOW = 6;
+/** How far forward to look for that retry's resolving (non-error) result. */
+export declare const ERROR_RESOLVE_WINDOW = 10;
 export declare const METRICS_BUFFER_MAX = 200;
 export declare const DAMAGE_LOOKBACK_MSGS = 15;
 export declare const VERIFICATION_GAP_SNIPPET_LEN = 80;

package/dist/constants.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"constants.d.ts","sourceRoot":"","sources":["../src/constants.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAEpE;;;;;GAKG;AACH,eAAO,MAAM,OAAO,WAAW,CAAC;AAChC,eAAO,MAAM,eAAe,MAAM,CAAC;AAEnC,eAAO,MAAM,qBAAqB,QAKqB,CAAC;AAExD,eAAO,MAAM,QAAQ,EAAE,MAAM,CAAC,kBAAkB,EAAE,aAAa,CAyB9D,CAAC;AAEF,eAAO,MAAM,cAAc;aACF,kBAAkB;;kBAEnB,MAAM,GAAG,IAAI;uBACR,MAAM,GAAG,IAAI;;;;;;cAMxB,MAAM,EAAE;CACzB,CAAC;AAEF,eAAO,MAAM,QAAQ,QAA+E,CAAC;AACrG,eAAO,MAAM,QAAQ,QAAuH,CAAC;AAC7I,eAAO,MAAM,SAAS,QAAmG,CAAC;AAI1H,eAAO,MAAM,kBAAkB,QAgB+C,CAAC;AAE/E,eAAO,MAAM,kBAAkB,yHACyF,CAAC;AAEzH,eAAO,MAAM,mBAAmB,QAWoB,CAAC;AAErD,eAAO,MAAM,mBAAmB,8DAA8D,CAAC;AAE/F,eAAO,MAAM,sBAAsB,QAoB2B,CAAC;AAE/D,eAAO,MAAM,sBAAsB,8OAC0M,CAAC;AAI9O,eAAO,MAAM,yBAAyB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAK5D,CAAC;AAGF,eAAO,MAAM,YAAY,YAAY,CAAC;AACtC,eAAO,MAAM,mBAAmB,iCAAiC,CAAC;AAClE,eAAO,MAAM,gBAAgB,gBAAgB,CAAC;AAC9C,eAAO,MAAM,iBAAiB,qBAAqB,CAAC;AACpD,eAAO,MAAM,sBAAsB,sBAAsB,CAAC;AAC1D,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,wBAAwB,wBAAwB,CAAC;AAC9D,eAAO,MAAM,cAAc,sBAAsB,CAAC;AAClD,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,eAAe,qCAAqC,CAAC;AAGlE,eAAO,MAAM,UAAU,oBAAoB,CAAC;AAG5C,eAAO,MAAM,mBAAmB,OAAO,CAAC;AACxC,eAAO,MAAM,sBAAsB,IAAI,CAAC;AAQxC,eAAO,MAAM,gBAAgB,KAAK,CAAC;AACnC,eAAO,MAAM,iBAAiB,QAA2B,CAAC;AAO1D,eAAO,MAAM,mBAAmB,MAAM,CAAC;AACvC,eAAO,MAAM,iBAAiB,KAAK,CAAC;AACpC,eAAO,MAAM,qBAAqB,KAAK,CAAC;AACxC,eAAO,MAAM,wBAAwB,KAAK,CAAC;AAG3C,eAAO,MAAM,qBAAqB,MAAM,CAAC;AAOzC,eAAO,MAAM,kBAAkB,MAAM,CAAC;AAGtC,eAAO,MAAM,oBAAoB,KAAK,CAAC;AAGvC,eAAO,MAAM,4BAA4B,KAAK,CAAC;AAG/C,eAAO,MAAM,UAAU,iBAAiB,CAAC;AACzC,eAAO,MAAM,cAAc,oBAAoB,CAAC;AAEhD,eAAO,MAAM,sBAAsB,QAU6T,CAAC"}
+{"version":3,"file":"constants.d.ts","sourceRoot":"","sources":["../src/constants.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,kBAAkB,EAAE,aAAa,EAAE,MAAM,YAAY,CAAC;AAEpE;;;;;GAKG;AACH,eAAO,MAAM,OAAO,WAAW,CAAC;AAChC,eAAO,MAAM,eAAe,MAAM,CAAC;AAEnC,eAAO,MAAM,qBAAqB,QAKqB,CAAC;AAExD,eAAO,MAAM,QAAQ,EAAE,MAAM,CAAC,kBAAkB,EAAE,aAAa,CAyB9D,CAAC;AAEF,eAAO,MAAM,cAAc;aACF,kBAAkB;;kBAEnB,MAAM,GAAG,IAAI;uBACR,MAAM,GAAG,IAAI;;;;;;cAMxB,MAAM,EAAE;CACzB,CAAC;AAEF,eAAO,MAAM,QAAQ,QAA+E,CAAC;AAIrG,eAAO,MAAM,QAAQ,QAAwK,CAAC;AAC9L,eAAO,MAAM,SAAS,QAAmG,CAAC;AAI1H,eAAO,MAAM,kBAAkB,QAgB+C,CAAC;AAE/E,eAAO,MAAM,kBAAkB,yHACyF,CAAC;AAEzH,eAAO,MAAM,mBAAmB,QAWoB,CAAC;AAErD,eAAO,MAAM,mBAAmB,8DAA8D,CAAC;AAE/F,eAAO,MAAM,sBAAsB,QAoB2B,CAAC;AAE/D,eAAO,MAAM,sBAAsB,8OAC0M,CAAC;AAI9O,eAAO,MAAM,yBAAyB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAK5D,CAAC;AAGF,eAAO,MAAM,YAAY,YAAY,CAAC;AACtC,eAAO,MAAM,mBAAmB,iCAAiC,CAAC;AAClE,eAAO,MAAM,gBAAgB,gBAAgB,CAAC;AAC9C,eAAO,MAAM,iBAAiB,qBAAqB,CAAC;AACpD,eAAO,MAAM,sBAAsB,sBAAsB,CAAC;AAC1D,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,wBAAwB,wBAAwB,CAAC;AAC9D,eAAO,MAAM,cAAc,sBAAsB,CAAC;AAClD,eAAO,MAAM,kBAAkB,kBAAkB,CAAC;AAClD,eAAO,MAAM,eAAe,qCAAqC,CAAC;AAGlE,eAAO,MAAM,UAAU,oBAAoB,CAAC;AAG5C,eAAO,MAAM,mBAAmB,OAAO,CAAC;AACxC,eAAO,MAAM,sBAAsB,IAAI,CAAC;AAQxC,eAAO,MAAM,gBAAgB,KAAK,CAAC;AACnC,eAAO,MAAM,iBAAiB,QAA2B,CAAC;AAG1D,eAAO,MAAM,eAAe,QAAgB,CAAC;AAC7C,eAAO,MAAM,WAAW,QAAiB,CAAC;AAC1C,eAAO,MAAM,aAAa,QAA0B,CAAC;AACrD,eAAO,MAAM,cAAc,QAA2B,CAAC;AAGvD,eAAO,MAAM,uBAAuB,wBAAwB,CAAC;AAG7D,eAAO,MAAM,SAAS;;;;;;;CAOZ,CAAC;AAGX,eAAO,MAAM,MAAM;;;;;;;;CAQT,CAAC;AAGX,eAAO,MAAM,KAAK;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8BR,CAAC;AAGX,eAAO,MAAM,wBAAwB,KAAK,CAAC;AAG3C,eAAO,MAAM,qBAAqB,MAAM,CAAC;AAGzC,sFAAsF;AACtF,eAAO,MAAM,eAAe,QAA8J,CAAC;AAC3L,+FAA+F;AAC/F,eAAO,MAAM,kBAAkB,OAAO,CAAC;AACvC,2EAA2E;AAC3E,eAAO,MAAM,kBAAkB,IAAI,CAAC;AACpC,6EAA6E;AAC7E,eAAO,MAAM,oBAAoB,KAAK,CAAC;AAOvC,eAAO,MAAM,kBAAkB,MAAM,CAAC;AAGtC,eAAO,MAAM,oBAAoB,KAAK,CAAC;AAGvC,eAAO,MAAM,4BAA4B,KAAK,CAAC;AAG/C,eAAO,MAAM,UAAU,iBAAiB,CAAC;AACzC,eAAO,MAAM,cAAc,oBAAoB,CAAC;AAEhD,eAAO,MAAM,sBAAsB,QAU6T,CAAC"}

package/dist/domain/keywords.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Salient-keyword extraction for fuzzy "is this concept covered?" checks.
+ *
+ * Pure (no I/O) — lives in the domain layer so both the verify phase and the
+ * damage detector share one implementation instead of each re-deriving a
+ * "first N long words" heuristic (which picked positional noise and missed the
+ * real term further into the text).
+ *
+ * ponytail ceiling: this is still keyword/substring matching. A paraphrase
+ * ("TS" vs "TypeScript") can't be recognized — an inherent limit of the cheap
+ * proxy, not fixable without an LLM judge (which would defeat deterministic-first).
+ */
+/**
+ * Extract salient keyword tokens from a source string. Strips punctuation and
+ * prefers proper nouns / identifiers (capitalized or digit-bearing) over
+ * positional fillers. Returns at most `max` tokens, original case preserved
+ * (callers lower-case as needed for comparison).
+ */
+export declare function extractCheckKeywords(text: string, max: number): string[];
+//# sourceMappingURL=keywords.d.ts.map

package/dist/domain/keywords.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"keywords.d.ts","sourceRoot":"","sources":["../../src/domain/keywords.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,MAAM,GAAG,MAAM,EAAE,CAQxE"}

package/dist/domain/tool-semantics.d.ts

@@ -0,0 +1,45 @@
+/**
+ * Name-agnostic tool classification by argument shape.
+ *
+ * Extraction used to classify tools by NAME — `hasToolHint(name, ["write",
+ * "edit", ...])` and `tc.name === "bash"`. That is a proxy: a tool is a write
+ * tool because its ARGUMENTS carry a content payload, not because its name
+ * happens to contain "write". Name matching is also churn-fragile —
+ * `read` → `hypa_read` → whatever-next silently drops off the list, and every
+ * new tool demands a code change.
+ *
+ * We classify by what the tool CARRIES, which is invariant under renaming:
+ *
+ *   mutates   — a path-like arg AND a content payload (write/edit/append/...)
+ *   executes  — a command/shell arg (bash/hypa_shell/...)
+ *   accesses  — a path-like arg with no payload (read/grep/find/ls)
+ *   other     — none of the above (ask_user, process, ...)
+ *
+ * The argument-name conventions (path/content/command) are far more stable
+ * across tools and providers than tool names, so this auto-adapts to tools
+ * this code has never seen (hypa_*, MCP tools, custom extensions) without a
+ * single name in a list.
+ *
+ * Ceiling (documented, not hidden): a DELETE cannot be told from a READ by
+ * arguments alone — both carry only a path. Callers that need deletion must
+ * resolve it from the result text. Bash-style tools that take a free-text
+ * `command` are opaque to file-path tracking by construction.
+ *
+ * Pure: no I/O, no async, no globals — lives in the domain layer alongside
+ * `summary-schema.ts`.
+ */
+/**
+ * Extract a non-empty file-path argument from a tool call, covering common key
+ * names. Returns undefined when no usable path is present.
+ */
+export declare function extractToolPath(args: unknown): string | undefined;
+export type ToolClass = "mutates" | "accesses" | "executes" | "other";
+/**
+ * Classify a tool by its argument shape, independent of its name.
+ *
+ * Order matters: a path + payload wins as `mutates` even if a command key is
+ * also present (a write tool that logs a command is still a write). A bare
+ * command with no path is `executes`. A bare path is `accesses`.
+ */
+export declare function classifyTool(args: unknown): ToolClass;
+//# sourceMappingURL=tool-semantics.d.ts.map

package/dist/domain/tool-semantics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tool-semantics.d.ts","sourceRoot":"","sources":["../../src/domain/tool-semantics.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA6BG;AA4BH;;;GAGG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,OAAO,GAAG,MAAM,GAAG,SAAS,CAQjE;AAED,MAAM,MAAM,SAAS,GAAG,SAAS,GAAG,UAAU,GAAG,UAAU,GAAG,OAAO,CAAC;AAEtE;;;;;;GAMG;AACH,wBAAgB,YAAY,CAAC,IAAI,EAAE,OAAO,GAAG,SAAS,CAUrD"}

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAoB,MAAM,iCAAiC,CAAC;AA4EtF,MAAM,CAAC,OAAO,UAAU,qBAAqB,CAAC,EAAE,EAAE,YAAY,QAsQ7D"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAoB,MAAM,iCAAiC,CAAC;AA0EtF,MAAM,CAAC,OAAO,UAAU,qBAAqB,CAAC,EAAE,EAAE,YAAY,QAsQ7D"}