cognitive-core 0.2.5 → 0.3.1

package/dist/eval/live/registry-store.js

@@ -0,0 +1,272 @@
+/**
+ * Registry storage abstraction — one interface, two backends — for the
+ * cross-run eval registries (difficulty + lift).
+ *
+ * Backends:
+ *   - SQLite (canonical default): a standalone long-lived DB file, by default
+ *     `eval-traces/registry.db`. Deliberately NOT the per-instance `.atlas`
+ *     database — eval Atlas instances are throwaway temp dirs, while the
+ *     registry must accumulate across runs. Same engineering discipline as
+ *     `persistence/sqlite-persistence.ts`: better-sqlite3, WAL, prepared
+ *     statements.
+ *   - JSON file (optional): the pre-existing difficulty-registry JSON format,
+ *     extended with an optional `lift` section. The original round-trip
+ *     functions in difficulty-registry.ts remain the implementation.
+ *
+ * The pure registry logic (estimateFamilySuccess, discriminativeWeight,
+ * selectDiscriminativeTasks, estimateLift, ingest helpers) stays
+ * backend-agnostic: stores load cells into the in-memory shapes those
+ * functions already consume (`DifficultyRegistryFile`, `LiftRegistryData`)
+ * and persist them back. Load → mutate (pure ingest) → save; save has
+ * last-writer-wins-per-cell semantics on both backends.
+ */
+import Database from 'better-sqlite3';
+import { existsSync, mkdirSync, readFileSync } from 'node:fs';
+import { dirname } from 'node:path';
+import { emptyDifficultyRegistry, loadDifficultyRegistry, saveDifficultyRegistry, } from './difficulty-registry.js';
+import { emptyLiftRegistry } from './lift-registry.js';
+/** Default canonical registry location (created on first write). */
+export const DEFAULT_REGISTRY_DB_PATH = 'eval-traces/registry.db';
+/** Scope key used for production (dream-time) lift rollups. */
+export const PRODUCTION_LIFT_SCOPE = 'production';
+/**
+ * JSON file backend. Difficulty round-trips through the original
+ * difficulty-registry functions (existing files stay loadable untouched);
+ * lift lives under an optional `lift` key in the same file. Each save
+ * re-reads the file first so difficulty writes never drop the lift section
+ * and vice versa.
+ */
+export class JsonRegistryStore {
+    path;
+    description;
+    constructor(path) {
+        this.path = path;
+        this.description = `JSON registry ${path}`;
+    }
+    readRaw() {
+        if (!existsSync(this.path))
+            return emptyDifficultyRegistry();
+        return JSON.parse(readFileSync(this.path, 'utf8'));
+    }
+    loadDifficulty() {
+        return loadDifficultyRegistry(this.path);
+    }
+    saveDifficulty(registry) {
+        const current = this.readRaw();
+        const merged = { ...registry };
+        if (current.lift && !registry.lift)
+            merged.lift = current.lift;
+        saveDifficultyRegistry(this.path, merged);
+    }
+    loadLift() {
+        const raw = this.readRaw();
+        return raw.lift ? { scopes: raw.lift } : emptyLiftRegistry();
+    }
+    saveLift(data) {
+        const current = this.readRaw();
+        const merged = { ...current, lift: data.scopes };
+        saveDifficultyRegistry(this.path, merged);
+    }
+    close() {
+        // Stateless — nothing to release.
+    }
+}
+/** DDL shared by the eval registry DB and the .atlas instance DB (persistence). */
+export const LIFT_CELLS_DDL = `
+  CREATE TABLE IF NOT EXISTS lift_cells (
+    scope              TEXT NOT NULL,
+    family             TEXT NOT NULL,
+    solver             TEXT NOT NULL,
+    guided_attempts    INTEGER NOT NULL DEFAULT 0,
+    guided_successes   INTEGER NOT NULL DEFAULT 0,
+    unguided_attempts  INTEGER NOT NULL DEFAULT 0,
+    unguided_successes INTEGER NOT NULL DEFAULT 0,
+    runs               INTEGER NOT NULL DEFAULT 0,
+    updated_at         TEXT NOT NULL,
+    PRIMARY KEY (scope, family, solver)
+  );
+`;
+const DIFFICULTY_CELLS_DDL = `
+  CREATE TABLE IF NOT EXISTS difficulty_cells (
+    benchmark      TEXT NOT NULL,
+    family         TEXT NOT NULL,
+    attempts       INTEGER NOT NULL DEFAULT 0,
+    successes      INTEGER NOT NULL DEFAULT 0,
+    cold_attempts  INTEGER NOT NULL DEFAULT 0,
+    cold_successes INTEGER NOT NULL DEFAULT 0,
+    runs           INTEGER NOT NULL DEFAULT 0,
+    updated_at     TEXT NOT NULL,
+    PRIMARY KEY (benchmark, family)
+  );
+`;
+/**
+ * SQLite backend (canonical default). WAL mode, prepared statements, cell
+ * upserts inside a single transaction per save.
+ */
+export class SqliteRegistryStore {
+    description;
+    db;
+    readonly;
+    stmts = {};
+    constructor(path, options = {}) {
+        this.readonly = options.readonly ?? false;
+        this.description = `SQLite registry ${path}`;
+        if (this.readonly) {
+            this.db = new Database(path, { readonly: true, fileMustExist: true });
+        }
+        else {
+            mkdirSync(dirname(path) || '.', { recursive: true });
+            this.db = new Database(path);
+            this.db.pragma('journal_mode = WAL');
+            this.db.pragma('busy_timeout = 5000');
+            this.db.exec(DIFFICULTY_CELLS_DDL + LIFT_CELLS_DDL);
+        }
+    }
+    stmt(name, sql) {
+        if (!this.stmts[name])
+            this.stmts[name] = this.db.prepare(sql);
+        return this.stmts[name];
+    }
+    hasTable(name) {
+        const row = this.db
+            .prepare(`SELECT name FROM sqlite_master WHERE type = 'table' AND name = ?`)
+            .get(name);
+        return row !== undefined;
+    }
+    loadDifficulty() {
+        const registry = emptyDifficultyRegistry();
+        if (!this.hasTable('difficulty_cells'))
+            return registry;
+        const rows = this.stmt('load_difficulty', 'SELECT * FROM difficulty_cells ORDER BY benchmark, family').all();
+        let latest = registry.updatedAt;
+        for (const row of rows) {
+            const bench = (registry.benchmarks[row.benchmark] ??= { families: {} });
+            const stats = {
+                attempts: row.attempts,
+                successes: row.successes,
+                coldAttempts: row.cold_attempts,
+                coldSuccesses: row.cold_successes,
+                runs: row.runs,
+            };
+            bench.families[row.family] = stats;
+            if (row.updated_at > latest)
+                latest = row.updated_at;
+        }
+        registry.updatedAt = latest;
+        return registry;
+    }
+    saveDifficulty(registry) {
+        const now = new Date().toISOString();
+        const upsert = this.stmt('upsert_difficulty', `
+      INSERT INTO difficulty_cells
+        (benchmark, family, attempts, successes, cold_attempts, cold_successes, runs, updated_at)
+      VALUES (@benchmark, @family, @attempts, @successes, @coldAttempts, @coldSuccesses, @runs, @now)
+      ON CONFLICT(benchmark, family) DO UPDATE SET
+        attempts = excluded.attempts,
+        successes = excluded.successes,
+        cold_attempts = excluded.cold_attempts,
+        cold_successes = excluded.cold_successes,
+        runs = excluded.runs,
+        updated_at = excluded.updated_at
+      `);
+        this.db.transaction(() => {
+            for (const [benchmark, { families }] of Object.entries(registry.benchmarks)) {
+                for (const [family, stats] of Object.entries(families)) {
+                    upsert.run({ benchmark, family, ...stats, now });
+                }
+            }
+        })();
+    }
+    loadLift() {
+        const data = emptyLiftRegistry();
+        if (!this.hasTable('lift_cells'))
+            return data;
+        const rows = this.stmt('load_lift', 'SELECT * FROM lift_cells ORDER BY scope, family, solver').all();
+        for (const row of rows) {
+            const scope = (data.scopes[row.scope] ??= { families: {} });
+            const solvers = (scope.families[row.family] ??= {});
+            solvers[row.solver] = {
+                guidedAttempts: row.guided_attempts,
+                guidedSuccesses: row.guided_successes,
+                unguidedAttempts: row.unguided_attempts,
+                unguidedSuccesses: row.unguided_successes,
+                runs: row.runs,
+            };
+        }
+        return data;
+    }
+    saveLift(data) {
+        const now = new Date().toISOString();
+        const upsert = this.stmt('upsert_lift', `
+      INSERT INTO lift_cells
+        (scope, family, solver, guided_attempts, guided_successes,
+         unguided_attempts, unguided_successes, runs, updated_at)
+      VALUES (@scope, @family, @solver, @guidedAttempts, @guidedSuccesses,
+              @unguidedAttempts, @unguidedSuccesses, @runs, @now)
+      ON CONFLICT(scope, family, solver) DO UPDATE SET
+        guided_attempts = excluded.guided_attempts,
+        guided_successes = excluded.guided_successes,
+        unguided_attempts = excluded.unguided_attempts,
+        unguided_successes = excluded.unguided_successes,
+        runs = excluded.runs,
+        updated_at = excluded.updated_at
+      `);
+        this.db.transaction(() => {
+            for (const [scope, { families }] of Object.entries(data.scopes)) {
+                for (const [family, solvers] of Object.entries(families)) {
+                    for (const [solver, cell] of Object.entries(solvers)) {
+                        upsert.run({ scope, family, solver, ...cell, now });
+                    }
+                }
+            }
+        })();
+    }
+    close() {
+        this.db.close();
+    }
+}
+/**
+ * Map CLI flags/config to a store instance. Precedence:
+ *   1. `registryDb`   → SQLite at that path.
+ *   2. `registryJson` → JSON file at that path (the pre-existing
+ *      --difficulty-registry contract, unchanged).
+ *   3. Neither        → SQLite at the canonical default path
+ *      (`eval-traces/registry.db`) — the default for the lift features.
+ */
+export function resolveRegistryStore(flags = {}) {
+    if (flags.registryDb)
+        return new SqliteRegistryStore(flags.registryDb);
+    if (flags.registryJson)
+        return new JsonRegistryStore(flags.registryJson);
+    return new SqliteRegistryStore(DEFAULT_REGISTRY_DB_PATH);
+}
+/**
+ * One-shot migration: copy a JSON registry file (difficulty + optional lift)
+ * into a SQLite registry DB. Idempotent — re-running overwrites the same
+ * cells with the same values. Returns cell counts for reporting.
+ */
+export function migrateJsonRegistryToSqlite(fromJsonPath, toDbPath) {
+    const source = new JsonRegistryStore(fromJsonPath);
+    const target = new SqliteRegistryStore(toDbPath);
+    try {
+        const difficulty = source.loadDifficulty();
+        const lift = source.loadLift();
+        target.saveDifficulty(difficulty);
+        target.saveLift(lift);
+        let difficultyCells = 0;
+        for (const { families } of Object.values(difficulty.benchmarks)) {
+            difficultyCells += Object.keys(families).length;
+        }
+        let liftCells = 0;
+        for (const { families } of Object.values(lift.scopes)) {
+            for (const solvers of Object.values(families)) {
+                liftCells += Object.keys(solvers).length;
+            }
+        }
+        return { difficultyCells, liftCells };
+    }
+    finally {
+        target.close();
+    }
+}
+//# sourceMappingURL=registry-store.js.map

package/dist/eval/live/registry-store.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"registry-store.js","sourceRoot":"","sources":["../../../src/eval/live/registry-store.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,OAAO,QAAQ,MAAM,gBAAgB,CAAC;AACtC,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAC9D,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EACL,uBAAuB,EACvB,sBAAsB,EACtB,sBAAsB,GAGvB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAE,iBAAiB,EAAyB,MAAM,oBAAoB,CAAC;AAE9E,oEAAoE;AACpE,MAAM,CAAC,MAAM,wBAAwB,GAAG,yBAAyB,CAAC;AAElE,+DAA+D;AAC/D,MAAM,CAAC,MAAM,qBAAqB,GAAG,YAAY,CAAC;AAqBlD;;;;;;GAMG;AACH,MAAM,OAAO,iBAAiB;IAGC;IAFpB,WAAW,CAAS;IAE7B,YAA6B,IAAY;QAAZ,SAAI,GAAJ,IAAI,CAAQ;QACvC,IAAI,CAAC,WAAW,GAAG,iBAAiB,IAAI,EAAE,CAAC;IAC7C,CAAC;IAEO,OAAO;QACb,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC;YAAE,OAAO,uBAAuB,EAAE,CAAC;QAC7D,OAAO,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,CAAqB,CAAC;IACzE,CAAC;IAED,cAAc;QACZ,OAAO,sBAAsB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC3C,CAAC;IAED,cAAc,CAAC,QAAgC;QAC7C,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAqB,EAAE,GAAG,QAAQ,EAAE,CAAC;QACjD,IAAI,OAAO,CAAC,IAAI,IAAI,CAAE,QAA6B,CAAC,IAAI;YAAE,MAAM,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC;QACrF,sBAAsB,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC5C,CAAC;IAED,QAAQ;QACN,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;QAC3B,OAAO,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,iBAAiB,EAAE,CAAC;IAC/D,CAAC;IAED,QAAQ,CAAC,IAAsB;QAC7B,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAqB,EAAE,GAAG,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;QACnE,sBAAsB,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IAC5C,CAAC;IAED,KAAK;QACH,kCAAkC;IACpC,CAAC;CACF;AA+BD,mFAAmF;AACnF,MAAM,CAAC,MAAM,cAAc,GAAG;;;;;;;;;;;;;CAa7B,CAAC;AAEF,MAAM,oBAAoB,GAAG;;;;;;;;;;;;CAY5B,CAAC;AAWF;;;GAGG;AACH,MAAM,OAAO,mBAAmB;IACrB,WAAW,CAAS;IACrB,EAAE,CAAoB;IACb,QAAQ,CAAU;IAC3B,KAAK,GAAuC,EAAE,CAAC;IAEvD,YAAY,IAAY,EAAE,UAAsC,EAAE;QAChE,IAAI,CAAC,QAAQ,GAAG,OAAO,CAAC,QAAQ,IAAI,KAAK,CAAC;QAC1C,IAAI,CAAC,WAAW,GAAG,mBAAmB,IAAI,EAAE,CAAC;QAC7C,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,IAAI,CAAC,EAAE,GAAG,IAAI,QAAQ,CAAC,IAAI,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;QACxE,CAAC;aAAM,CAAC;YACN,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACrD,IAAI,CAAC,EAAE,GAAG,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC;YAC7B,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,oBAAoB,CAAC,CAAC;YACrC,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,qBAAqB,CAAC,CAAC;YACtC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,oBAAoB,GAAG,cAAc,CAAC,CAAC;QACtD,CAAC;IACH,CAAC;IAEO,IAAI,CAAC,IAAY,EAAE,GAAW;QACpC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;YAAE,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QAC/D,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAEO,QAAQ,CAAC,IAAY;QAC3B,MAAM,GAAG,GAAG,IAAI,CAAC,EAAE;aAChB,OAAO,CAAC,kEAAkE,CAAC;aAC3E,GAAG,CAAC,IAAI,CAAC,CAAC;QACb,OAAO,GAAG,KAAK,SAAS,CAAC;IAC3B,CAAC;IAED,cAAc;QACZ,MAAM,QAAQ,GAAG,uBAAuB,EAAE,CAAC;QAC3C,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,kBAAkB,CAAC;YAAE,OAAO,QAAQ,CAAC;QACxD,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CACpB,iBAAiB,EACjB,2DAA2D,CAC5D,CAAC,GAAG,EAAyB,CAAC;QAC/B,IAAI,MAAM,GAAG,QAAQ,CAAC,SAAS,CAAC;QAChC,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,KAAK,GAAG,CAAC,QAAQ,CAAC,UAAU,CAAC,GAAG,CAAC,SAAS,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;YACxE,MAAM,KAAK,GAA0B;gBACnC,QAAQ,EAAE,GAAG,CAAC,QAAQ;gBACtB,SAAS,EAAE,GAAG,CAAC,SAAS;gBACxB,YAAY,EAAE,GAAG,CAAC,aAAa;gBAC/B,aAAa,EAAE,GAAG,CAAC,cAAc;gBACjC,IAAI,EAAE,GAAG,CAAC,IAAI;aACf,CAAC;YACF,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,KAAK,CAAC;YACnC,IAAI,GAAG,CAAC,UAAU,GAAG,MAAM;gBAAE,MAAM,GAAG,GAAG,CAAC,UAAU,CAAC;QACvD,CAAC;QACD,QAAQ,CAAC,SAAS,GAAG,MAAM,CAAC;QAC5B,OAAO,QAAQ,CAAC;IAClB,CAAC;IAED,cAAc,CAAC,QAAgC;QAC7C,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QACrC,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CACtB,mBAAmB,EACnB;;;;;;;;;;;OAWC,CACF,CAAC;QACF,IAAI,CAAC,EAAE,CAAC,WAAW,CAAC,GAAG,EAAE;YACvB,KAAK,MAAM,CAAC,SAAS,EAAE,EAAE,QAAQ,EAAE,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC;gBAC5E,KAAK,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACvD,MAAM,CAAC,GAAG,CAAC,EAAE,SAAS,EAAE,MAAM,EAAE,GAAG,KAAK,EAAE,GAAG,EAAE,CAAC,CAAC;gBACnD,CAAC;YACH,CAAC;QACH,CAAC,CAAC,EAAE,CAAC;IACP,CAAC;IAED,QAAQ;QACN,MAAM,IAAI,GAAG,iBAAiB,EAAE,CAAC;QACjC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,YAAY,CAAC;YAAE,OAAO,IAAI,CAAC;QAC9C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CACpB,WAAW,EACX,yDAAyD,CAC1D,CAAC,GAAG,EAAmB,CAAC;QACzB,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;YAC5D,MAAM,OAAO,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,MAAM,CAAC,KAAK,EAAE,CAAC,CAAC;YACpD,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG;gBACpB,cAAc,EAAE,GAAG,CAAC,eAAe;gBACnC,eAAe,EAAE,GAAG,CAAC,gBAAgB;gBACrC,gBAAgB,EAAE,GAAG,CAAC,iBAAiB;gBACvC,iBAAiB,EAAE,GAAG,CAAC,kBAAkB;gBACzC,IAAI,EAAE,GAAG,CAAC,IAAI;aACf,CAAC;QACJ,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,QAAQ,CAAC,IAAsB;QAC7B,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QACrC,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CACtB,aAAa,EACb;;;;;;;;;;;;;OAaC,CACF,CAAC;QACF,IAAI,CAAC,EAAE,CAAC,WAAW,CAAC,GAAG,EAAE;YACvB,KAAK,MAAM,CAAC,KAAK,EAAE,EAAE,QAAQ,EAAE,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;gBAChE,KAAK,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;oBACzD,KAAK,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;wBACrD,MAAM,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,IAAI,EAAE,GAAG,EAAE,CAAC,CAAC;oBACtD,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC,CAAC,EAAE,CAAC;IACP,CAAC;IAED,KAAK;QACH,IAAI,CAAC,EAAE,CAAC,KAAK,EAAE,CAAC;IAClB,CAAC;CACF;AAaD;;;;;;;GAOG;AACH,MAAM,UAAU,oBAAoB,CAAC,QAA4B,EAAE;IACjE,IAAI,KAAK,CAAC,UAAU;QAAE,OAAO,IAAI,mBAAmB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IACvE,IAAI,KAAK,CAAC,YAAY;QAAE,OAAO,IAAI,iBAAiB,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;IACzE,OAAO,IAAI,mBAAmB,CAAC,wBAAwB,CAAC,CAAC;AAC3D,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,2BAA2B,CACzC,YAAoB,EACpB,QAAgB;IAEhB,MAAM,MAAM,GAAG,IAAI,iBAAiB,CAAC,YAAY,CAAC,CAAC;IACnD,MAAM,MAAM,GAAG,IAAI,mBAAmB,CAAC,QAAQ,CAAC,CAAC;IACjD,IAAI,CAAC;QACH,MAAM,UAAU,GAAG,MAAM,CAAC,cAAc,EAAE,CAAC;QAC3C,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,EAAE,CAAC;QAC/B,MAAM,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC;QAClC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACtB,IAAI,eAAe,GAAG,CAAC,CAAC;QACxB,KAAK,MAAM,EAAE,QAAQ,EAAE,IAAI,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,UAAU,CAAC,EAAE,CAAC;YAChE,eAAe,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;QAClD,CAAC;QACD,IAAI,SAAS,GAAG,CAAC,CAAC;QAClB,KAAK,MAAM,EAAE,QAAQ,EAAE,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;YACtD,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC9C,SAAS,IAAI,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;YAC3C,CAAC;QACH,CAAC;QACD,OAAO,EAAE,eAAe,EAAE,SAAS,EAAE,CAAC;IACxC,CAAC;YAAS,CAAC;QACT,MAAM,CAAC,KAAK,EAAE,CAAC;IACjB,CAAC;AACH,CAAC"}

package/dist/eval/live/run-checkpoint.d.ts

@@ -0,0 +1,55 @@
+import type { TaskResult } from '../types.js';
+/**
+ * Callback fired by LiveEvalRunner after each task completes — used for
+ * progress reporting and checkpointing. Defined here (below live-runner in
+ * the dependency DAG) because live-runner also imports cold-reuse.ts, which
+ * consumes CheckpointFile from this module; defining the hook in live-runner
+ * would close a cycle.
+ */
+export type TaskBoundaryHook = (event: {
+    arm: string;
+    seed: number;
+    taskIndex: number;
+    taskCount: number;
+    result: TaskResult;
+}) => void | Promise<void>;
+/** Per-arm running tally, recomputed on every checkpoint write. */
+export interface CheckpointArmSummary {
+    completed: number;
+    passed: number;
+    envErrors: number;
+}
+interface CheckpointTaskRow extends TaskResult {
+    arm: string;
+    seed: number;
+    taskCount: number;
+}
+/** Shape of the checkpoint JSON on disk. */
+export interface CheckpointFile {
+    partial: true;
+    meta: Record<string, unknown>;
+    startedAt: string;
+    updatedAt: string;
+    summary: Record<string, CheckpointArmSummary>;
+    tasks: CheckpointTaskRow[];
+}
+export interface RunCheckpoint {
+    /** Wire as (or into) LiveRunConfig.onTaskBoundary. */
+    record: TaskBoundaryHook;
+    /** Delete the checkpoint after the final report is safely written. */
+    finalize(): void;
+    /** Checkpoint file location (for logging). */
+    path: string;
+}
+/** `docs/runs/foo.json` → `docs/runs/foo.checkpoint.json`. */
+export declare function checkpointPathFor(outPath: string): string;
+/**
+ * `docs/runs/foo.json` → `docs/runs/foo.workdir`. Stable, deterministic working
+ * directory that sits next to the checkpoint so `--resume` (same `--out`) finds
+ * both the checkpoint AND the persisted learning-arm Atlas dirs
+ * (`<workdir>/learn-seed-<seed>`). Ignored by `.gitignore` (`*.workdir/`).
+ */
+export declare function workdirPathFor(outPath: string): string;
+export declare function createRunCheckpoint(outPath: string, meta: Record<string, unknown>): RunCheckpoint;
+export {};
+//# sourceMappingURL=run-checkpoint.d.ts.map

package/dist/eval/live/run-checkpoint.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-checkpoint.d.ts","sourceRoot":"","sources":["../../../src/eval/live/run-checkpoint.ts"],"names":[],"mappings":"AAkBA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAE9C;;;;;;GAMG;AACH,MAAM,MAAM,gBAAgB,GAAG,CAAC,KAAK,EAAE;IACrC,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,UAAU,CAAC;CACpB,KAAK,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;AAE3B,mEAAmE;AACnE,MAAM,WAAW,oBAAoB;IACnC,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,UAAU,iBAAkB,SAAQ,UAAU;IAC5C,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,4CAA4C;AAC5C,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,IAAI,CAAC;IACd,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC9B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,oBAAoB,CAAC,CAAC;IAC9C,KAAK,EAAE,iBAAiB,EAAE,CAAC;CAC5B;AAED,MAAM,WAAW,aAAa;IAC5B,sDAAsD;IACtD,MAAM,EAAE,gBAAgB,CAAC;IACzB,sEAAsE;IACtE,QAAQ,IAAI,IAAI,CAAC;IACjB,8CAA8C;IAC9C,IAAI,EAAE,MAAM,CAAC;CACd;AAED,8DAA8D;AAC9D,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAIzD;AAED;;;;;GAKG;AACH,wBAAgB,cAAc,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAItD;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,aAAa,CAgDjG"}

package/dist/eval/live/run-checkpoint.js

@@ -0,0 +1,86 @@
+/**
+ * Incremental run checkpointing for live evals.
+ *
+ * Live runs are hours long and every episode costs real money; the final
+ * report JSON is only assembled at the very end, so a crash / reaped process
+ * used to lose the aggregate view of everything already paid for (per-task
+ * traces survive, but reassembling them is manual). The checkpoint file is a
+ * running partial result written after EVERY completed task: if the run dies,
+ * the checkpoint holds per-task results and per-arm summaries up to the last
+ * completed episode.
+ *
+ * Written atomically (temp file + rename) so a kill mid-write can never leave
+ * a truncated JSON. Deleted on successful completion of the final report —
+ * the presence of a `<name>.checkpoint.json` therefore always means "this run
+ * did not finish".
+ */
+import { mkdirSync, renameSync, rmSync, writeFileSync } from 'node:fs';
+import { dirname } from 'node:path';
+/** `docs/runs/foo.json` → `docs/runs/foo.checkpoint.json`. */
+export function checkpointPathFor(outPath) {
+    return outPath.endsWith('.json')
+        ? `${outPath.slice(0, -'.json'.length)}.checkpoint.json`
+        : `${outPath}.checkpoint.json`;
+}
+/**
+ * `docs/runs/foo.json` → `docs/runs/foo.workdir`. Stable, deterministic working
+ * directory that sits next to the checkpoint so `--resume` (same `--out`) finds
+ * both the checkpoint AND the persisted learning-arm Atlas dirs
+ * (`<workdir>/learn-seed-<seed>`). Ignored by `.gitignore` (`*.workdir/`).
+ */
+export function workdirPathFor(outPath) {
+    return outPath.endsWith('.json')
+        ? `${outPath.slice(0, -'.json'.length)}.workdir`
+        : `${outPath}.workdir`;
+}
+export function createRunCheckpoint(outPath, meta) {
+    const path = checkpointPathFor(outPath);
+    const startedAt = new Date().toISOString();
+    const tasks = [];
+    const record = (event) => {
+        tasks.push({
+            arm: event.arm,
+            seed: event.seed,
+            taskCount: event.taskCount,
+            ...event.result,
+        });
+        const summary = {};
+        for (const t of tasks) {
+            const s = (summary[t.arm] ??= { completed: 0, passed: 0, envErrors: 0 });
+            s.completed++;
+            if (t.status === 'env_error')
+                s.envErrors++;
+            else if (t.success)
+                s.passed++;
+        }
+        const body = {
+            partial: true,
+            meta,
+            startedAt,
+            updatedAt: new Date().toISOString(),
+            summary,
+            tasks,
+        };
+        // Checkpointing must never take down the run it protects.
+        try {
+            mkdirSync(dirname(path), { recursive: true });
+            const tmp = `${path}.tmp`;
+            writeFileSync(tmp, JSON.stringify(body, null, 2));
+            renameSync(tmp, path);
+        }
+        catch {
+            // Non-fatal: disk hiccups lose one checkpoint tick, not the run.
+        }
+    };
+    const finalize = () => {
+        try {
+            rmSync(path, { force: true });
+            rmSync(`${path}.tmp`, { force: true });
+        }
+        catch {
+            // Non-fatal.
+        }
+    };
+    return { record, finalize, path };
+}
+//# sourceMappingURL=run-checkpoint.js.map

package/dist/eval/live/run-checkpoint.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-checkpoint.js","sourceRoot":"","sources":["../../../src/eval/live/run-checkpoint.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AACH,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACvE,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAkDpC,8DAA8D;AAC9D,MAAM,UAAU,iBAAiB,CAAC,OAAe;IAC/C,OAAO,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;QAC9B,CAAC,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,kBAAkB;QACxD,CAAC,CAAC,GAAG,OAAO,kBAAkB,CAAC;AACnC,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,cAAc,CAAC,OAAe;IAC5C,OAAO,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;QAC9B,CAAC,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,UAAU;QAChD,CAAC,CAAC,GAAG,OAAO,UAAU,CAAC;AAC3B,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,OAAe,EAAE,IAA6B;IAChF,MAAM,IAAI,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;IACxC,MAAM,SAAS,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAC3C,MAAM,KAAK,GAAwB,EAAE,CAAC;IAEtC,MAAM,MAAM,GAAqB,CAAC,KAAK,EAAE,EAAE;QACzC,KAAK,CAAC,IAAI,CAAC;YACT,GAAG,EAAE,KAAK,CAAC,GAAG;YACd,IAAI,EAAE,KAAK,CAAC,IAAI;YAChB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,GAAG,KAAK,CAAC,MAAM;SAChB,CAAC,CAAC;QACH,MAAM,OAAO,GAAyC,EAAE,CAAC;QACzD,KAAK,MAAM,CAAC,IAAI,KAAK,EAAE,CAAC;YACtB,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,CAAC,CAAC;YACzE,CAAC,CAAC,SAAS,EAAE,CAAC;YACd,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW;gBAAE,CAAC,CAAC,SAAS,EAAE,CAAC;iBACvC,IAAI,CAAC,CAAC,OAAO;gBAAE,CAAC,CAAC,MAAM,EAAE,CAAC;QACjC,CAAC;QACD,MAAM,IAAI,GAAmB;YAC3B,OAAO,EAAE,IAAI;YACb,IAAI;YACJ,SAAS;YACT,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,OAAO;YACP,KAAK;SACN,CAAC;QACF,0DAA0D;QAC1D,IAAI,CAAC;YACH,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YAC9C,MAAM,GAAG,GAAG,GAAG,IAAI,MAAM,CAAC;YAC1B,aAAa,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;YAClD,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QACxB,CAAC;QAAC,MAAM,CAAC;YACP,iEAAiE;QACnE,CAAC;IACH,CAAC,CAAC;IAEF,MAAM,QAAQ,GAAG,GAAS,EAAE;QAC1B,IAAI,CAAC;YACH,MAAM,CAAC,IAAI,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YAC9B,MAAM,CAAC,GAAG,IAAI,MAAM,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACzC,CAAC;QAAC,MAAM,CAAC;YACP,aAAa;QACf,CAAC;IACH,CAAC,CAAC;IAEF,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;AACpC,CAAC"}

package/dist/eval/live/run-resume.d.ts

@@ -0,0 +1,22 @@
+import type { TaskResult } from '../types.js';
+/** Adopted per-(arm, seed) completed task results loaded from a checkpoint. */
+export interface ResumeState {
+    /** Meta block copied from the checkpoint (benchmark, model, seeds, …). */
+    readonly meta: Record<string, unknown>;
+    /** ISO timestamp the aborted run started (for logging). */
+    readonly startedAt?: string;
+    /** Total completed task rows loaded across all arms/seeds. */
+    readonly totalCompleted: number;
+    /**
+     * Completed results for one arm+seed, keyed by the task's stream index.
+     * Undefined when the aborted run recorded nothing for that arm+seed.
+     */
+    completedFor(arm: string, seed: number): Map<number, TaskResult> | undefined;
+}
+/**
+ * Load resume state from a `<out>.checkpoint.json` partial file. Throws when the
+ * path is missing or is not a partial checkpoint — silently resuming from a
+ * final/empty artifact would re-run the whole stream and quietly waste the run.
+ */
+export declare function loadResumeState(path: string): ResumeState;
+//# sourceMappingURL=run-resume.d.ts.map

package/dist/eval/live/run-resume.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-resume.d.ts","sourceRoot":"","sources":["../../../src/eval/live/run-resume.ts"],"names":[],"mappings":"AAkBA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAG9C,+EAA+E;AAC/E,MAAM,WAAW,WAAW;IAC1B,0EAA0E;IAC1E,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACvC,2DAA2D;IAC3D,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,8DAA8D;IAC9D,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC;;;OAGG;IACH,YAAY,CAAC,GAAG,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,GAAG,CAAC,MAAM,EAAE,UAAU,CAAC,GAAG,SAAS,CAAC;CAC9E;AAmBD;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAiCzD"}

package/dist/eval/live/run-resume.js

@@ -0,0 +1,70 @@
+/**
+ * Resume state for aborted live-eval runs.
+ *
+ * A live run writes a `<out>.checkpoint.json` partial file after every completed
+ * task (see run-checkpoint.ts). When a run dies (workspace/process-tree
+ * teardown), that checkpoint holds every episode already paid for. Resuming
+ * reconstructs an uninterrupted run by (a) adopting the already-completed
+ * `TaskResult`s per (arm, seed, index) instead of re-solving them, and (b) — for
+ * the learning arm — reopening the STABLE persisted Atlas `baseDir`
+ * (`<out>.workdir/learn-seed-<seed>`) so accumulated playbooks/experiences carry
+ * forward. The aggregated final report is therefore identical to a run that
+ * never died.
+ *
+ * This module owns only the read side: turning a checkpoint file into a
+ * per-(arm, seed) index of completed task results. The runner (live-runner.ts)
+ * consumes it to skip completed indices and reopen learning memory.
+ */
+import { existsSync, readFileSync } from 'node:fs';
+/** Strip a checkpoint row down to a plain `TaskResult` (drops arm/seed/taskCount). */
+function rowToTaskResult(row) {
+    return {
+        index: row.index,
+        taskId: row.taskId,
+        skillTag: row.skillTag,
+        success: row.success,
+        status: row.status,
+        stepCount: row.stepCount ?? 0,
+        knowledgeSurfaced: row.knowledgeSurfaced ?? false,
+        priorSkillCoverage: row.priorSkillCoverage ?? 0,
+        wallTimeMs: row.wallTimeMs ?? 0,
+        ...(typeof row.costUsd === 'number' ? { costUsd: row.costUsd } : {}),
+        ...(typeof row.partialScore === 'number' ? { partialScore: row.partialScore } : {}),
+    };
+}
+/**
+ * Load resume state from a `<out>.checkpoint.json` partial file. Throws when the
+ * path is missing or is not a partial checkpoint — silently resuming from a
+ * final/empty artifact would re-run the whole stream and quietly waste the run.
+ */
+export function loadResumeState(path) {
+    if (!existsSync(path)) {
+        throw new Error(`Resume checkpoint not found: ${path}`);
+    }
+    const raw = JSON.parse(readFileSync(path, 'utf8'));
+    if (raw.partial !== true) {
+        throw new Error(`Resume artifact ${path} is not a partial checkpoint (partial !== true). ` +
+            `Only <out>.checkpoint.json files written by an in-flight run are resumable.`);
+    }
+    const file = raw;
+    // arm -> seed -> index -> TaskResult
+    const byArm = new Map();
+    let totalCompleted = 0;
+    for (const row of file.tasks ?? []) {
+        const bySeed = byArm.get(row.arm) ?? new Map();
+        byArm.set(row.arm, bySeed);
+        const byIndex = bySeed.get(row.seed) ?? new Map();
+        bySeed.set(row.seed, byIndex);
+        // Last write wins if a row repeats (checkpoints are append-only per task, so
+        // duplicates only arise across resume generations — the newest is correct).
+        byIndex.set(row.index, rowToTaskResult(row));
+        totalCompleted++;
+    }
+    return {
+        meta: (file.meta ?? {}),
+        startedAt: file.startedAt,
+        totalCompleted,
+        completedFor: (arm, seed) => byArm.get(arm)?.get(seed),
+    };
+}
+//# sourceMappingURL=run-resume.js.map

package/dist/eval/live/run-resume.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run-resume.js","sourceRoot":"","sources":["../../../src/eval/live/run-resume.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AACH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAmBnD,sFAAsF;AACtF,SAAS,eAAe,CAAC,GAA4B;IACnD,OAAO;QACL,KAAK,EAAE,GAAG,CAAC,KAAe;QAC1B,MAAM,EAAE,GAAG,CAAC,MAAgB;QAC5B,QAAQ,EAAE,GAAG,CAAC,QAAkB;QAChC,OAAO,EAAE,GAAG,CAAC,OAAkB;QAC/B,MAAM,EAAE,GAAG,CAAC,MAA8B;QAC1C,SAAS,EAAG,GAAG,CAAC,SAAoB,IAAI,CAAC;QACzC,iBAAiB,EAAG,GAAG,CAAC,iBAA6B,IAAI,KAAK;QAC9D,kBAAkB,EAAG,GAAG,CAAC,kBAA6B,IAAI,CAAC;QAC3D,UAAU,EAAG,GAAG,CAAC,UAAqB,IAAI,CAAC;QAC3C,GAAG,CAAC,OAAO,GAAG,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACpE,GAAG,CAAC,OAAO,GAAG,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,GAAG,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpF,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAAC,IAAY;IAC1C,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,gCAAgC,IAAI,EAAE,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAA4B,CAAC;IAC9E,IAAI,GAAG,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;QACzB,MAAM,IAAI,KAAK,CACb,mBAAmB,IAAI,mDAAmD;YACxE,6EAA6E,CAChF,CAAC;IACJ,CAAC;IAED,MAAM,IAAI,GAAG,GAAgC,CAAC;IAC9C,qCAAqC;IACrC,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;IACtE,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,KAAK,MAAM,GAAG,IAAI,IAAI,CAAC,KAAK,IAAI,EAAE,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,GAAG,EAAmC,CAAC;QAChF,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC3B,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,IAAI,GAAG,EAAsB,CAAC;QACtE,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC9B,6EAA6E;QAC7E,4EAA4E;QAC5E,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,EAAE,eAAe,CAAC,GAAyC,CAAC,CAAC,CAAC;QACnF,cAAc,EAAE,CAAC;IACnB,CAAC;IAED,OAAO;QACL,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,IAAI,EAAE,CAA4B;QAClD,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,cAAc;QACd,YAAY,EAAE,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,IAAI,CAAC;KACvD,CAAC;AACJ,CAAC"}

package/dist/eval/live/skill-injector.d.ts

@@ -0,0 +1,36 @@
+import type { Atlas } from '../../atlas.js';
+import type { Task, Playbook } from '../../types/index.js';
+import type { DockerOrchestrator } from './docker.js';
+/**
+ * Export playbooks relevant to a task as SKILL.md files into a Docker container.
+ *
+ * Uses the existing convertPlaybookToSkill() from skill-publisher.ts to produce
+ * the skill-tree Skill format, then renders it as a SKILL.md file that
+ * SkillsBench (and similar benchmarks) can consume.
+ */
+export interface SkillInjectionResult {
+    playbooksExported: number;
+    skillFiles: string[];
+}
+export declare function injectPlaybooksAsSkills(opts: {
+    atlas: Atlas;
+    task: Task;
+    docker: DockerOrchestrator;
+    containerId: string;
+    containerSkillDir: string;
+}): Promise<SkillInjectionResult>;
+/**
+ * Find playbooks matching a task without going through the full SkillLibrary
+ * tier system. Useful when Atlas doesn't have a SkillLibrary configured.
+ */
+export declare function findMatchingPlaybooks(opts: {
+    atlas: Atlas;
+    task: Task;
+    maxCount?: number;
+}): Promise<Playbook[]>;
+/**
+ * Export playbooks to SKILL.md files in a local directory (no Docker needed).
+ * Returns paths of created files.
+ */
+export declare function exportPlaybooksToDir(playbooks: Playbook[], outputDir: string): string[];
+//# sourceMappingURL=skill-injector.d.ts.map

package/dist/eval/live/skill-injector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"skill-injector.d.ts","sourceRoot":"","sources":["../../../src/eval/live/skill-injector.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,gBAAgB,CAAC;AAC5C,OAAO,KAAK,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AAE3D,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAEtD;;;;;;GAMG;AAEH,MAAM,WAAW,oBAAoB;IACnC,iBAAiB,EAAE,MAAM,CAAC;IAC1B,UAAU,EAAE,MAAM,EAAE,CAAC;CACtB;AAED,wBAAsB,uBAAuB,CAAC,IAAI,EAAE;IAClD,KAAK,EAAE,KAAK,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,MAAM,EAAE,kBAAkB,CAAC;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB,EAAE,MAAM,CAAC;CAC3B,GAAG,OAAO,CAAC,oBAAoB,CAAC,CA6BhC;AAiBD;;;GAGG;AACH,wBAAsB,qBAAqB,CAAC,IAAI,EAAE;IAChD,KAAK,EAAE,KAAK,CAAC;IACb,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB,GAAG,OAAO,CAAC,QAAQ,EAAE,CAAC,CAOtB;AAED;;;GAGG;AACH,wBAAgB,oBAAoB,CAAC,SAAS,EAAE,QAAQ,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,MAAM,EAAE,CAWvF"}

package/dist/eval/live/skill-injector.js

@@ -0,0 +1,81 @@
+import { writeFileSync, mkdtempSync, rmSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { convertPlaybookToSkill } from '../../surfacing/skill-publisher.js';
+export async function injectPlaybooksAsSkills(opts) {
+    const { atlas, task, docker, containerId, containerSkillDir } = opts;
+    const skills = await atlas.getSkillLibrary()?.getSkillsForAgent(task);
+    if (!skills)
+        return { playbooksExported: 0, skillFiles: [] };
+    const playbooks = [...skills.core, ...skills.domain, ...skills.contextual];
+    if (playbooks.length === 0)
+        return { playbooksExported: 0, skillFiles: [] };
+    const tmpDir = mkdtempSync(join(tmpdir(), 'cc-skill-inject-'));
+    const skillFiles = [];
+    try {
+        for (const playbook of playbooks) {
+            const skill = convertPlaybookToSkill(playbook);
+            const md = renderSkillMd(skill.name, skill.description, skill.instructions);
+            const filename = sanitizeFilename(playbook.name) + '.md';
+            const hostPath = join(tmpDir, filename);
+            writeFileSync(hostPath, md, 'utf-8');
+            const containerPath = `${containerSkillDir}/${filename}`;
+            await docker.copyToContainer(containerId, hostPath, containerPath);
+            skillFiles.push(containerPath);
+        }
+    }
+    finally {
+        rmSync(tmpDir, { recursive: true, force: true });
+    }
+    return { playbooksExported: playbooks.length, skillFiles };
+}
+/**
+ * Render a skill as a SKILL.md markdown file matching the Claude Code skill format.
+ */
+function renderSkillMd(name, description, instructions) {
+    return [
+        `---`,
+        `name: ${name}`,
+        `description: ${description.slice(0, 1024)}`,
+        `---`,
+        '',
+        instructions,
+        '',
+    ].join('\n');
+}
+/**
+ * Find playbooks matching a task without going through the full SkillLibrary
+ * tier system. Useful when Atlas doesn't have a SkillLibrary configured.
+ */
+export async function findMatchingPlaybooks(opts) {
+    const memory = opts.atlas.getMemory();
+    const matches = await memory.playbooks.findMatching(opts.task.description, {
+        minScore: 0.3,
+        k: opts.maxCount ?? 5,
+    });
+    return matches.map((m) => m.playbook);
+}
+/**
+ * Export playbooks to SKILL.md files in a local directory (no Docker needed).
+ * Returns paths of created files.
+ */
+export function exportPlaybooksToDir(playbooks, outputDir) {
+    const paths = [];
+    for (const playbook of playbooks) {
+        const skill = convertPlaybookToSkill(playbook);
+        const md = renderSkillMd(skill.name, skill.description, skill.instructions);
+        const filename = sanitizeFilename(playbook.name) + '.md';
+        const filepath = join(outputDir, filename);
+        writeFileSync(filepath, md, 'utf-8');
+        paths.push(filepath);
+    }
+    return paths;
+}
+function sanitizeFilename(name) {
+    return (name
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, '-')
+        .replace(/^-|-$/g, '')
+        .slice(0, 80) || 'skill');
+}
+//# sourceMappingURL=skill-injector.js.map