@astragenie/astramemory-local 0.7.2

package/dist/contracts/memory.js

@@ -0,0 +1,5 @@
+/** Single runtime source of truth for memory types — CLI validation and any
+ *  future enum checks must use this array, not a hand-copied list (the CLI
+ *  once drifted to a stale 5-type copy and silently rejected note/event). */
+export const MEMORY_TYPES = ['decision', 'fact', 'lesson', 'command', 'todo', 'note', 'event'];
+//# sourceMappingURL=memory.js.map

package/dist/contracts/memory.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"memory.js","sourceRoot":"","sources":["../../src/contracts/memory.ts"],"names":[],"mappings":"AAAA;;6EAE6E;AAC7E,MAAM,CAAC,MAAM,YAAY,GAAG,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAU,CAAC"}

package/dist/contracts/vector.d.ts

@@ -0,0 +1,29 @@
+/**
+ * Filters applied to vector search. Implementations MUST throw if they cannot
+ * honor a non-empty filter — silent ignore is a contract violation.
+ */
+export interface VecFilter {
+    type?: string[];
+    repo?: string;
+    project?: string;
+    since?: number;
+}
+export interface VecHit {
+    id: string;
+    score: number;
+}
+/**
+ * Vector store contract. v1 ships sqlite-vec only. LanceDB lands in a later
+ * wave behind the same interface; add 'lancedb' to the union when that
+ * implementation exists.
+ */
+export interface VectorStore {
+    readonly name: 'sqlite-vec';
+    upsert(id: string, vec: Float32Array): Promise<void>;
+    search(vec: Float32Array, k: number, filter?: VecFilter): Promise<VecHit[]>;
+    /**
+     * Erase all stored vectors. Used as the wipe primitive ahead of a
+     * re-embed pass orchestrated by a Wave 3 worker. Does NOT repopulate.
+     */
+    clear(): Promise<void>;
+}

package/dist/contracts/vector.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=vector.js.map

package/dist/contracts/vector.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"vector.js","sourceRoot":"","sources":["../../src/contracts/vector.ts"],"names":[],"mappings":""}

package/dist/distill/flatten-turns.d.ts

@@ -0,0 +1 @@
+export declare function flattenTranscriptContent(content: string): string;

package/dist/distill/flatten-turns.js

@@ -0,0 +1,50 @@
+import { z } from 'zod';
+/**
+ * D-DEF1 fix — turn-flattening defect.
+ *
+ * The canonical ingest envelope (server/routes/ingest.ts) stores turns as
+ * `JSON.stringify(turns)` in `transcripts.content` — a structured
+ * `[{role, text, ts}]` array. The distill pipeline's stages 1-3
+ * (cleanup/normalize/chunk) were built for plain text with `role: text`
+ * line-prefixed turn boundaries — the same convention the legacy envelope's
+ * raw `content` string already uses (see distill/stages/03-chunk.ts
+ * `TURN_BOUNDARY_RE`). `JSON.stringify` produces a single-line string (no
+ * embedded newlines), so when that raw JSON is fed through unchanged, stage
+ * 3's turn-boundary detector never fires: the whole transcript becomes one
+ * opaque "turn" that gets sliced blindly by character count, and role
+ * attribution is lost before stage 1 even runs (CHANGELOG 0.2.0 "Known
+ * issues" — D-DEF1: "fix is scoped to... flatten turns to `role: text\n`
+ * pairs before pipeline entry").
+ *
+ * This converts the canonical JSON-turns shape into `role: text` pairs (one
+ * per line) — the format the pipeline stages already parse. Legacy
+ * plain-text content (or any JSON that isn't a turns array) passes through
+ * unchanged.
+ */
+const TurnLikeSchema = z.object({
+    role: z.string(),
+    text: z.string(),
+});
+export function flattenTranscriptContent(content) {
+    let parsed;
+    try {
+        parsed = JSON.parse(content);
+    }
+    catch {
+        return content; // legacy plain text — not JSON, pass through unchanged
+    }
+    if (!Array.isArray(parsed) || parsed.length === 0) {
+        return content;
+    }
+    const turns = [];
+    for (const item of parsed) {
+        const result = TurnLikeSchema.safeParse(item);
+        if (!result.success) {
+            // Not a turns array (some other JSON payload) — leave untouched.
+            return content;
+        }
+        turns.push(result.data);
+    }
+    return turns.map(t => `${t.role}: ${t.text}`).join('\n');
+}
+//# sourceMappingURL=flatten-turns.js.map

package/dist/distill/flatten-turns.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"flatten-turns.js","sourceRoot":"","sources":["../../src/distill/flatten-turns.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9B,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;CACjB,CAAC,CAAC;AAEH,MAAM,UAAU,wBAAwB,CAAC,OAAe;IACtD,IAAI,MAAe,CAAC;IACpB,IAAI,CAAC;QACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;IAC/B,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,OAAO,CAAC,CAAC,uDAAuD;IACzE,CAAC;IAED,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClD,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,MAAM,KAAK,GAA0C,EAAE,CAAC;IACxD,KAAK,MAAM,IAAI,IAAI,MAAM,EAAE,CAAC;QAC1B,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;QAC9C,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,iEAAiE;YACjE,OAAO,OAAO,CAAC;QACjB,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1B,CAAC;IAED,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC3D,CAAC"}

package/dist/distill/pipeline.d.ts

@@ -0,0 +1,45 @@
+/**
+ * 8-stage distillation pipeline.
+ *
+ * Stages:
+ * 1. cleanup     — deterministic
+ * 2. normalize   — deterministic
+ * 3. chunk       — deterministic
+ * 4. compact     — LLM call (budget-gated)
+ * 5. extract     — LLM call, JSON-mode (budget-gated, Zod-validated, retry once)
+ * 6. reduce      — deterministic
+ * 7. memory-normalize — deterministic
+ * 8. embed + index    — provider call + DB writes
+ */
+import type { LLMProvider, EmbedProvider } from '../contracts/index.js';
+import type { DB } from '../storage/db.js';
+export { BudgetExceeded } from '../budget/tracker.js';
+export interface PipelineContext {
+    db: DB;
+    llmCompaction: LLMProvider;
+    llmExtraction: LLMProvider;
+    embed: EmbedProvider;
+    capUsd: number;
+    sessionId: string | null;
+    repo: string | null;
+    project: string | null;
+    branch: string | null;
+    agent: string | null;
+    sourceHash: string | null;
+}
+export interface PipelineResult {
+    memoriesCreated: number;
+    memoriesDeduped: number;
+    atomsExtracted: number;
+    chunksProcessed: number;
+    /** IDs of every memory this run touched (both newly-created and deduped-existing), in stage-8 order. */
+    memoryIds: string[];
+}
+/**
+ * Run the full 8-stage distillation pipeline on a raw transcript.
+ * Returns count of new memories created.
+ *
+ * Throws BudgetExceeded if the daily cap is hit during stages 4 or 5.
+ * The caller (distill handler) catches this and moves the job to paused.
+ */
+export declare function runPipeline(transcript: string, ctx: PipelineContext): Promise<PipelineResult>;

package/dist/distill/pipeline.js

@@ -0,0 +1,113 @@
+/**
+ * 8-stage distillation pipeline.
+ *
+ * Stages:
+ * 1. cleanup     — deterministic
+ * 2. normalize   — deterministic
+ * 3. chunk       — deterministic
+ * 4. compact     — LLM call (budget-gated)
+ * 5. extract     — LLM call, JSON-mode (budget-gated, Zod-validated, retry once)
+ * 6. reduce      — deterministic
+ * 7. memory-normalize — deterministic
+ * 8. embed + index    — provider call + DB writes
+ */
+import { cleanup } from './stages/01-cleanup.js';
+import { normalize } from './stages/02-normalize.js';
+import { chunk } from './stages/03-chunk.js';
+import { compact, BudgetExceeded as CompactBudgetExceeded } from './stages/04-compact.js';
+import { extract, BudgetExceeded as ExtractBudgetExceeded } from './stages/05-extract.js';
+import { reduce } from './stages/06-reduce.js';
+import { memoryNormalize } from './stages/07-memory-normalize.js';
+import { embedAndIndex } from './stages/08-embed-index.js';
+import { BudgetTracker } from '../budget/tracker.js';
+import { childLogger } from '../log/logger.js';
+export { BudgetExceeded } from '../budget/tracker.js';
+/**
+ * Run the full 8-stage distillation pipeline on a raw transcript.
+ * Returns count of new memories created.
+ *
+ * Throws BudgetExceeded if the daily cap is hit during stages 4 or 5.
+ * The caller (distill handler) catches this and moves the job to paused.
+ */
+export async function runPipeline(transcript, ctx) {
+    // Base logger for the pipeline run — inherits any job_id from caller context
+    const pipeLog = childLogger({ session_id: ctx.sessionId ?? undefined });
+    // Stage 1: Cleanup
+    pipeLog.info({ stage: 'cleanup' }, 'stage start');
+    const cleaned = cleanup(transcript);
+    pipeLog.info({ stage: 'cleanup' }, 'stage complete');
+    // Stage 2: Normalize
+    pipeLog.info({ stage: 'normalize' }, 'stage start');
+    const normalized = normalize(cleaned);
+    pipeLog.info({ stage: 'normalize' }, 'stage complete');
+    // Stage 3: Chunk
+    pipeLog.info({ stage: 'chunk' }, 'stage start');
+    const chunks = chunk(normalized);
+    pipeLog.info({ stage: 'chunk', chunk_count: chunks.length }, 'stage complete');
+    const budget = new BudgetTracker(ctx.db);
+    // Stage 4: Compact (LLM, budget-gated)
+    pipeLog.info({ stage: 'compact', chunk_count: chunks.length }, 'stage start');
+    let compactedChunks;
+    try {
+        const compactResults = await compact(chunks, ctx.llmCompaction, budget, ctx.capUsd);
+        compactedChunks = compactResults.map(r => ({ index: r.index, text: r.text || chunks[r.index]?.text || '' }));
+        pipeLog.info({ stage: 'compact', chunks_compacted: compactedChunks.length }, 'stage complete');
+    }
+    catch (err) {
+        if (err instanceof CompactBudgetExceeded) {
+            pipeLog.warn({ stage: 'compact', error_kind: 'BudgetExceeded' }, 'stage aborted — budget exceeded');
+            throw err;
+        }
+        pipeLog.warn({ stage: 'compact', error_message: (err instanceof Error ? err.message : String(err)).slice(0, 200), error_kind: err instanceof Error ? err.name : 'Unknown' }, 'stage failed');
+        throw err;
+    }
+    // Stage 5: Extract (LLM JSON-mode, budget-gated, retry on parse fail)
+    pipeLog.info({ stage: 'extract', chunk_count: compactedChunks.length }, 'stage start');
+    let extractResults;
+    try {
+        extractResults = await extract(compactedChunks, ctx.llmExtraction, budget, ctx.capUsd);
+        const atomsExtracted = extractResults.reduce((sum, r) => sum + r.atoms.length, 0);
+        pipeLog.info({ stage: 'extract', atoms_extracted: atomsExtracted }, 'stage complete');
+    }
+    catch (err) {
+        if (err instanceof ExtractBudgetExceeded) {
+            pipeLog.warn({ stage: 'extract', error_kind: 'BudgetExceeded' }, 'stage aborted — budget exceeded');
+            throw err;
+        }
+        pipeLog.warn({ stage: 'extract', error_message: (err instanceof Error ? err.message : String(err)).slice(0, 200), error_kind: err instanceof Error ? err.name : 'Unknown' }, 'stage failed');
+        throw err;
+    }
+    // Flatten all atoms
+    const allAtoms = extractResults.flatMap(r => r.atoms);
+    // Stage 6: Reduce (merge by content hash)
+    pipeLog.info({ stage: 'reduce', atoms_in: allAtoms.length }, 'stage start');
+    const reducedAtoms = reduce(allAtoms);
+    pipeLog.info({ stage: 'reduce', atoms_out: reducedAtoms.length }, 'stage complete');
+    // Stage 7: Memory-normalize
+    pipeLog.info({ stage: 'memory-normalize' }, 'stage start');
+    const normalizedMemories = memoryNormalize(reducedAtoms);
+    pipeLog.info({ stage: 'memory-normalize', memories: normalizedMemories.length }, 'stage complete');
+    // Stage 8: Embed + index
+    pipeLog.info({ stage: 'embed-index', memories: normalizedMemories.length }, 'stage start');
+    const indexResults = await embedAndIndex(normalizedMemories, {
+        db: ctx.db,
+        embed: ctx.embed,
+        sessionId: ctx.sessionId,
+        repo: ctx.repo,
+        project: ctx.project,
+        branch: ctx.branch,
+        agent: ctx.agent,
+        sourceHash: ctx.sourceHash,
+    });
+    const created = indexResults.filter(r => r.created).length;
+    const deduped = indexResults.filter(r => !r.created).length;
+    pipeLog.info({ stage: 'embed-index', memories_created: created, memories_deduped: deduped }, 'stage complete');
+    return {
+        memoriesCreated: created,
+        memoriesDeduped: deduped,
+        atomsExtracted: allAtoms.length,
+        chunksProcessed: chunks.length,
+        memoryIds: indexResults.map(r => r.memoryId),
+    };
+}
+//# sourceMappingURL=pipeline.js.map

package/dist/distill/pipeline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pipeline.js","sourceRoot":"","sources":["../../src/distill/pipeline.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,wBAAwB,CAAC;AACjD,OAAO,EAAE,SAAS,EAAE,MAAM,0BAA0B,CAAC;AACrD,OAAO,EAAE,KAAK,EAAE,MAAM,sBAAsB,CAAC;AAC7C,OAAO,EAAE,OAAO,EAAE,cAAc,IAAI,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC1F,OAAO,EAAE,OAAO,EAAE,cAAc,IAAI,qBAAqB,EAAE,MAAM,wBAAwB,CAAC;AAC1F,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAC/C,OAAO,EAAE,eAAe,EAAE,MAAM,iCAAiC,CAAC;AAClE,OAAO,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAC3D,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAGrD,OAAO,EAAE,WAAW,EAAE,MAAM,kBAAkB,CAAC;AAE/C,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAyBtD;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAC/B,UAAkB,EAClB,GAAoB;IAEpB,6EAA6E;IAC7E,MAAM,OAAO,GAAG,WAAW,CAAC,EAAE,UAAU,EAAE,GAAG,CAAC,SAAS,IAAI,SAAS,EAAE,CAAC,CAAC;IAExE,mBAAmB;IACnB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,EAAE,aAAa,CAAC,CAAC;IAClD,MAAM,OAAO,GAAG,OAAO,CAAC,UAAU,CAAC,CAAC;IACpC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAErD,qBAAqB;IACrB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,EAAE,aAAa,CAAC,CAAC;IACpD,MAAM,UAAU,GAAG,SAAS,CAAC,OAAO,CAAC,CAAC;IACtC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,WAAW,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAEvD,iBAAiB;IACjB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE,aAAa,CAAC,CAAC;IAChD,MAAM,MAAM,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC;IACjC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAE/E,MAAM,MAAM,GAAG,IAAI,aAAa,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAEzC,uCAAuC;IACvC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC;IAC9E,IAAI,eAAuD,CAAC;IAC5D,IAAI,CAAC;QACH,MAAM,cAAc,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,aAAa,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;QACpF,eAAe,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,IAAI,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,IAAI,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC;QAC7G,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,gBAAgB,EAAE,eAAe,CAAC,MAAM,EAAE,EAAE,gBAAgB,CAAC,CAAC;IACjG,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,qBAAqB,EAAE,CAAC;YACzC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,UAAU,EAAE,gBAAgB,EAAE,EAAE,iCAAiC,CAAC,CAAC;YACpG,MAAM,GAAG,CAAC;QACZ,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,aAAa,EAAE,CAAC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,UAAU,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAAE,EAAE,cAAc,CAAC,CAAC;QAC7L,MAAM,GAAG,CAAC;IACZ,CAAC;IAED,sEAAsE;IACtE,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,eAAe,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC;IACvF,IAAI,cAAc,CAAC;IACnB,IAAI,CAAC;QACH,cAAc,GAAG,MAAM,OAAO,CAAC,eAAe,EAAE,GAAG,CAAC,aAAa,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;QACvF,MAAM,cAAc,GAAG,cAAc,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAClF,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,cAAc,EAAE,EAAE,gBAAgB,CAAC,CAAC;IACxF,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,GAAG,YAAY,qBAAqB,EAAE,CAAC;YACzC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,UAAU,EAAE,gBAAgB,EAAE,EAAE,iCAAiC,CAAC,CAAC;YACpG,MAAM,GAAG,CAAC;QACZ,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,SAAS,EAAE,aAAa,EAAE,CAAC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,UAAU,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,EAAE,EAAE,cAAc,CAAC,CAAC;QAC7L,MAAM,GAAG,CAAC;IACZ,CAAC;IAED,oBAAoB;IACpB,MAAM,QAAQ,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC;IAEtD,0CAA0C;IAC1C,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC;IAC5E,MAAM,YAAY,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC;IACtC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,YAAY,CAAC,MAAM,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAEpF,4BAA4B;IAC5B,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,kBAAkB,EAAE,EAAE,aAAa,CAAC,CAAC;IAC3D,MAAM,kBAAkB,GAAG,eAAe,CAAC,YAAY,CAAC,CAAC;IACzD,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,kBAAkB,EAAE,QAAQ,EAAE,kBAAkB,CAAC,MAAM,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAEnG,yBAAyB;IACzB,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,aAAa,EAAE,QAAQ,EAAE,kBAAkB,CAAC,MAAM,EAAE,EAAE,aAAa,CAAC,CAAC;IAC3F,MAAM,YAAY,GAAG,MAAM,aAAa,CAAC,kBAAkB,EAAE;QAC3D,EAAE,EAAE,GAAG,CAAC,EAAE;QACV,KAAK,EAAE,GAAG,CAAC,KAAK;QAChB,SAAS,EAAE,GAAG,CAAC,SAAS;QACxB,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,OAAO,EAAE,GAAG,CAAC,OAAO;QACpB,MAAM,EAAE,GAAG,CAAC,MAAM;QAClB,KAAK,EAAE,GAAG,CAAC,KAAK;QAChB,UAAU,EAAE,GAAG,CAAC,UAAU;KAC3B,CAAC,CAAC;IAEH,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;IAC3D,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC;IAE5D,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,aAAa,EAAE,gBAAgB,EAAE,OAAO,EAAE,gBAAgB,EAAE,OAAO,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAE/G,OAAO;QACL,eAAe,EAAE,OAAO;QACxB,eAAe,EAAE,OAAO;QACxB,cAAc,EAAE,QAAQ,CAAC,MAAM;QAC/B,eAAe,EAAE,MAAM,CAAC,MAAM;QAC9B,SAAS,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC;KAC7C,CAAC;AACJ,CAAC"}

package/dist/distill/prompts/extract.d.ts

@@ -0,0 +1,122 @@
+import { z } from 'zod';
+export declare const EntityKindSchema: z.ZodEnum<["product", "tech", "person", "org", "repo", "concept"]>;
+export type EntityKind = z.infer<typeof EntityKindSchema>;
+export declare const AtomEntitySchema: z.ZodObject<{
+    name: z.ZodString;
+    kind: z.ZodEnum<["product", "tech", "person", "org", "repo", "concept"]>;
+}, "strip", z.ZodTypeAny, {
+    kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+    name: string;
+}, {
+    kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+    name: string;
+}>;
+export type AtomEntity = z.infer<typeof AtomEntitySchema>;
+export declare const AtomSchema: z.ZodObject<{
+    type: z.ZodEnum<["decision", "fact", "lesson", "command", "todo", "note", "event"]>;
+    text: z.ZodString;
+    importance: z.ZodNumber;
+    confidence: z.ZodNumber;
+    evidence: z.ZodOptional<z.ZodString>;
+    /** Named things mentioned in this atom (FEAT-402 AC-1). Optional — atoms
+     * with no clearly-named entity omit it entirely rather than guessing. */
+    entities: z.ZodOptional<z.ZodArray<z.ZodObject<{
+        name: z.ZodString;
+        kind: z.ZodEnum<["product", "tech", "person", "org", "repo", "concept"]>;
+    }, "strip", z.ZodTypeAny, {
+        kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+        name: string;
+    }, {
+        kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+        name: string;
+    }>, "many">>;
+}, "strip", z.ZodTypeAny, {
+    type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+    text: string;
+    importance: number;
+    confidence: number;
+    evidence?: string | undefined;
+    entities?: {
+        kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+        name: string;
+    }[] | undefined;
+}, {
+    type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+    text: string;
+    importance: number;
+    confidence: number;
+    evidence?: string | undefined;
+    entities?: {
+        kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+        name: string;
+    }[] | undefined;
+}>;
+export declare const ExtractionSchema: z.ZodObject<{
+    atoms: z.ZodArray<z.ZodObject<{
+        type: z.ZodEnum<["decision", "fact", "lesson", "command", "todo", "note", "event"]>;
+        text: z.ZodString;
+        importance: z.ZodNumber;
+        confidence: z.ZodNumber;
+        evidence: z.ZodOptional<z.ZodString>;
+        /** Named things mentioned in this atom (FEAT-402 AC-1). Optional — atoms
+         * with no clearly-named entity omit it entirely rather than guessing. */
+        entities: z.ZodOptional<z.ZodArray<z.ZodObject<{
+            name: z.ZodString;
+            kind: z.ZodEnum<["product", "tech", "person", "org", "repo", "concept"]>;
+        }, "strip", z.ZodTypeAny, {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }, {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }>, "many">>;
+    }, "strip", z.ZodTypeAny, {
+        type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+        text: string;
+        importance: number;
+        confidence: number;
+        evidence?: string | undefined;
+        entities?: {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }[] | undefined;
+    }, {
+        type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+        text: string;
+        importance: number;
+        confidence: number;
+        evidence?: string | undefined;
+        entities?: {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }[] | undefined;
+    }>, "many">;
+}, "strip", z.ZodTypeAny, {
+    atoms: {
+        type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+        text: string;
+        importance: number;
+        confidence: number;
+        evidence?: string | undefined;
+        entities?: {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }[] | undefined;
+    }[];
+}, {
+    atoms: {
+        type: "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event";
+        text: string;
+        importance: number;
+        confidence: number;
+        evidence?: string | undefined;
+        entities?: {
+            kind: "org" | "product" | "tech" | "person" | "repo" | "concept";
+            name: string;
+        }[] | undefined;
+    }[];
+}>;
+export type Atom = z.infer<typeof AtomSchema>;
+export type Extraction = z.infer<typeof ExtractionSchema>;
+export declare const EXTRACTION_SYSTEM_PROMPT = "You extract durable knowledge from AI coding agent transcripts.\n\nOutput STRICT JSON matching:\n{\n  \"atoms\": [\n    {\n      \"type\": \"decision\" | \"fact\" | \"lesson\" | \"command\" | \"todo\" | \"note\" | \"event\",\n      \"text\": \"<5-500 chars, canonical phrasing>\",\n      \"importance\": 0.0-1.0,\n      \"confidence\": 0.0-1.0,\n      \"evidence\": \"<optional, short excerpt from source>\",\n      \"entities\": [{ \"name\": \"<named thing mentioned>\", \"kind\": \"product\" | \"tech\" | \"person\" | \"org\" | \"repo\" | \"concept\" }]\n    }\n  ]\n}\n\nTypes:\n- decision: chosen approach/architecture (use SQLite, not Postgres)\n- fact: discovered truth about the codebase/system (port 7777 is the default)\n- lesson: learning from failure or surprise (Bun does not support better-sqlite3 on Windows)\n- command: useful shell incantation (npm test -- --reporter=verbose)\n- todo: pending work item explicitly raised\n- note: freeform observation or annotation not fitting other types\n- event: something that happened at a point in time (deployment, incident, decision meeting)\n\nEntities: list every specific named thing the atom is ABOUT (not every word\nmentioned) \u2014 a technology, product, repo, person, or organization. kind is\none of:\n- product: a named product or app (e.g. \"AstraMemory\", \"GitHub Copilot\")\n- tech: a language/framework/library/tool/protocol (e.g. \"Postgres\", \"TypeScript\", \"sqlite-vec\")\n- person: a named individual\n- org: a company or organization\n- repo: a named git repository or codebase\n- concept: a named pattern, architecture, or idea (e.g. \"bitemporal model\", \"CRDT\")\nOmit \"entities\" entirely (or leave it empty) when the atom names nothing specific.\n\nSkip noise: greetings, repeated thinking, false starts, generic AI fluff.\nSkip secrets: tokens, API keys, passwords.\nConfidence reflects evidence strength: clear conclusion = 0.9; speculation = 0.3.\nImportance reflects long-term value: architecture decision = 0.9; one-off command = 0.4.";
+export declare const EXTRACTION_STRICT_PROMPT = "You extract durable knowledge from AI coding agent transcripts.\n\nOutput STRICT JSON matching:\n{\n  \"atoms\": [\n    {\n      \"type\": \"decision\" | \"fact\" | \"lesson\" | \"command\" | \"todo\" | \"note\" | \"event\",\n      \"text\": \"<5-500 chars, canonical phrasing>\",\n      \"importance\": 0.0-1.0,\n      \"confidence\": 0.0-1.0,\n      \"evidence\": \"<optional, short excerpt from source>\",\n      \"entities\": [{ \"name\": \"<named thing mentioned>\", \"kind\": \"product\" | \"tech\" | \"person\" | \"org\" | \"repo\" | \"concept\" }]\n    }\n  ]\n}\n\nTypes:\n- decision: chosen approach/architecture (use SQLite, not Postgres)\n- fact: discovered truth about the codebase/system (port 7777 is the default)\n- lesson: learning from failure or surprise (Bun does not support better-sqlite3 on Windows)\n- command: useful shell incantation (npm test -- --reporter=verbose)\n- todo: pending work item explicitly raised\n- note: freeform observation or annotation not fitting other types\n- event: something that happened at a point in time (deployment, incident, decision meeting)\n\nEntities: list every specific named thing the atom is ABOUT (not every word\nmentioned) \u2014 a technology, product, repo, person, or organization. kind is\none of:\n- product: a named product or app (e.g. \"AstraMemory\", \"GitHub Copilot\")\n- tech: a language/framework/library/tool/protocol (e.g. \"Postgres\", \"TypeScript\", \"sqlite-vec\")\n- person: a named individual\n- org: a company or organization\n- repo: a named git repository or codebase\n- concept: a named pattern, architecture, or idea (e.g. \"bitemporal model\", \"CRDT\")\nOmit \"entities\" entirely (or leave it empty) when the atom names nothing specific.\n\nSkip noise: greetings, repeated thinking, false starts, generic AI fluff.\nSkip secrets: tokens, API keys, passwords.\nConfidence reflects evidence strength: clear conclusion = 0.9; speculation = 0.3.\nImportance reflects long-term value: architecture decision = 0.9; one-off command = 0.4.\n\nCRITICAL: Output ONLY valid JSON. No markdown, no prose, no code fences. The response must start with { and end with }.";

package/dist/distill/prompts/extract.js

@@ -0,0 +1,67 @@
+import { z } from 'zod';
+// FEAT-402 (W4.2) AC-1/AC-9 — entity kinds emitted alongside each atom, in
+// the SAME extraction LLM call (no extra call/budget spend). Names are
+// normalized at write time via src/entity/normalize.ts, not here — this
+// schema only validates shape.
+export const EntityKindSchema = z.enum(['product', 'tech', 'person', 'org', 'repo', 'concept']);
+export const AtomEntitySchema = z.object({
+    name: z.string().min(1).max(100),
+    kind: EntityKindSchema,
+});
+export const AtomSchema = z.object({
+    type: z.enum(['decision', 'fact', 'lesson', 'command', 'todo', 'note', 'event']),
+    text: z.string().min(5).max(500),
+    importance: z.number().min(0).max(1),
+    confidence: z.number().min(0).max(1),
+    evidence: z.string().optional(),
+    /** Named things mentioned in this atom (FEAT-402 AC-1). Optional — atoms
+     * with no clearly-named entity omit it entirely rather than guessing. */
+    entities: z.array(AtomEntitySchema).optional(),
+});
+export const ExtractionSchema = z.object({
+    atoms: z.array(AtomSchema)
+});
+export const EXTRACTION_SYSTEM_PROMPT = `You extract durable knowledge from AI coding agent transcripts.
+
+Output STRICT JSON matching:
+{
+  "atoms": [
+    {
+      "type": "decision" | "fact" | "lesson" | "command" | "todo" | "note" | "event",
+      "text": "<5-500 chars, canonical phrasing>",
+      "importance": 0.0-1.0,
+      "confidence": 0.0-1.0,
+      "evidence": "<optional, short excerpt from source>",
+      "entities": [{ "name": "<named thing mentioned>", "kind": "product" | "tech" | "person" | "org" | "repo" | "concept" }]
+    }
+  ]
+}
+
+Types:
+- decision: chosen approach/architecture (use SQLite, not Postgres)
+- fact: discovered truth about the codebase/system (port 7777 is the default)
+- lesson: learning from failure or surprise (Bun does not support better-sqlite3 on Windows)
+- command: useful shell incantation (npm test -- --reporter=verbose)
+- todo: pending work item explicitly raised
+- note: freeform observation or annotation not fitting other types
+- event: something that happened at a point in time (deployment, incident, decision meeting)
+
+Entities: list every specific named thing the atom is ABOUT (not every word
+mentioned) — a technology, product, repo, person, or organization. kind is
+one of:
+- product: a named product or app (e.g. "AstraMemory", "GitHub Copilot")
+- tech: a language/framework/library/tool/protocol (e.g. "Postgres", "TypeScript", "sqlite-vec")
+- person: a named individual
+- org: a company or organization
+- repo: a named git repository or codebase
+- concept: a named pattern, architecture, or idea (e.g. "bitemporal model", "CRDT")
+Omit "entities" entirely (or leave it empty) when the atom names nothing specific.
+
+Skip noise: greetings, repeated thinking, false starts, generic AI fluff.
+Skip secrets: tokens, API keys, passwords.
+Confidence reflects evidence strength: clear conclusion = 0.9; speculation = 0.3.
+Importance reflects long-term value: architecture decision = 0.9; one-off command = 0.4.`;
+export const EXTRACTION_STRICT_PROMPT = `${EXTRACTION_SYSTEM_PROMPT}
+
+CRITICAL: Output ONLY valid JSON. No markdown, no prose, no code fences. The response must start with { and end with }.`;
+//# sourceMappingURL=extract.js.map

package/dist/distill/prompts/extract.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"extract.js","sourceRoot":"","sources":["../../../src/distill/prompts/extract.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,2EAA2E;AAC3E,uEAAuE;AACvE,wEAAwE;AACxE,+BAA+B;AAC/B,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC;AAGhG,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IACvC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAChC,IAAI,EAAE,gBAAgB;CACvB,CAAC,CAAC;AAGH,MAAM,CAAC,MAAM,UAAU,GAAG,CAAC,CAAC,MAAM,CAAC;IACjC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,UAAU,EAAE,MAAM,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;IAChF,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAChC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B;6EACyE;IACzE,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,QAAQ,EAAE;CAC/C,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,gBAAgB,GAAG,CAAC,CAAC,MAAM,CAAC;IACvC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,UAAU,CAAC;CAC3B,CAAC,CAAC;AAKH,MAAM,CAAC,MAAM,wBAAwB,GAAG;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;yFAuCiD,CAAC;AAE1F,MAAM,CAAC,MAAM,wBAAwB,GAAG,GAAG,wBAAwB;;wHAEqD,CAAC"}

package/dist/distill/stages/01-cleanup.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Stage 1 — Cleanup (deterministic)
+ *
+ * - Normalize line endings to \n
+ * - Deduplicate consecutive blank lines (max 2)
+ * - Remove repeated tool-output blocks (identical lines appearing 3+ times consecutively)
+ * - Strip trailing whitespace per line
+ */
+export declare function cleanup(text: string): string;

package/dist/distill/stages/01-cleanup.js

@@ -0,0 +1,67 @@
+/**
+ * Stage 1 — Cleanup (deterministic)
+ *
+ * - Normalize line endings to \n
+ * - Deduplicate consecutive blank lines (max 2)
+ * - Remove repeated tool-output blocks (identical lines appearing 3+ times consecutively)
+ * - Strip trailing whitespace per line
+ */
+/** Maximum consecutive blank lines to keep */
+const MAX_BLANK_LINES = 2;
+/** Number of identical consecutive lines that triggers dedup */
+const REPEAT_THRESHOLD = 3;
+export function cleanup(text) {
+    // 1. Normalize line endings
+    let result = text.replace(/\r\n/g, '\n').replace(/\r/g, '\n');
+    // 2. Strip trailing whitespace per line
+    result = result
+        .split('\n')
+        .map(line => line.trimEnd())
+        .join('\n');
+    // 3. Deduplicate repeated consecutive identical lines (tool output spam)
+    result = deduplicateRepeatedLines(result);
+    // 4. Collapse excess blank lines
+    result = collapseBlankLines(result);
+    // 5. Trim leading/trailing whitespace
+    return result.trim();
+}
+/**
+ * If the same non-empty line appears REPEAT_THRESHOLD or more times in a row,
+ * collapse them to a single instance.
+ */
+function deduplicateRepeatedLines(text) {
+    const lines = text.split('\n');
+    const out = [];
+    let i = 0;
+    while (i < lines.length) {
+        const line = lines[i] ?? '';
+        if (line.trim() === '') {
+            out.push(line);
+            i++;
+            continue;
+        }
+        // Count consecutive identical lines
+        let count = 1;
+        while (i + count < lines.length && lines[i + count] === line) {
+            count++;
+        }
+        if (count >= REPEAT_THRESHOLD) {
+            // Replace run with single occurrence + indicator
+            out.push(line);
+        }
+        else {
+            for (let j = 0; j < count; j++)
+                out.push(line);
+        }
+        i += count;
+    }
+    return out.join('\n');
+}
+/**
+ * Collapse runs of more than MAX_BLANK_LINES consecutive blank lines.
+ */
+function collapseBlankLines(text) {
+    // Replace 3+ consecutive newlines with 2 newlines (= MAX_BLANK_LINES blank line)
+    return text.replace(/\n{3,}/g, '\n\n');
+}
+//# sourceMappingURL=01-cleanup.js.map

package/dist/distill/stages/01-cleanup.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"01-cleanup.js","sourceRoot":"","sources":["../../../src/distill/stages/01-cleanup.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,8CAA8C;AAC9C,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,gEAAgE;AAChE,MAAM,gBAAgB,GAAG,CAAC,CAAC;AAE3B,MAAM,UAAU,OAAO,CAAC,IAAY;IAClC,4BAA4B;IAC5B,IAAI,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;IAE9D,wCAAwC;IACxC,MAAM,GAAG,MAAM;SACZ,KAAK,CAAC,IAAI,CAAC;SACX,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;SAC3B,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,yEAAyE;IACzE,MAAM,GAAG,wBAAwB,CAAC,MAAM,CAAC,CAAC;IAE1C,iCAAiC;IACjC,MAAM,GAAG,kBAAkB,CAAC,MAAM,CAAC,CAAC;IAEpC,sCAAsC;IACtC,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;AACvB,CAAC;AAED;;;GAGG;AACH,SAAS,wBAAwB,CAAC,IAAY;IAC5C,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,IAAI,CAAC,GAAG,CAAC,CAAC;IAEV,OAAO,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QAC5B,IAAI,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACvB,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACf,CAAC,EAAE,CAAC;YACJ,SAAS;QACX,CAAC;QAED,oCAAoC;QACpC,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,OAAO,CAAC,GAAG,KAAK,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,CAAC,CAAC,GAAG,KAAK,CAAC,KAAK,IAAI,EAAE,CAAC;YAC7D,KAAK,EAAE,CAAC;QACV,CAAC;QAED,IAAI,KAAK,IAAI,gBAAgB,EAAE,CAAC;YAC9B,iDAAiD;YACjD,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjB,CAAC;aAAM,CAAC;YACN,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE;gBAAE,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACjD,CAAC;QACD,CAAC,IAAI,KAAK,CAAC;IACb,CAAC;IAED,OAAO,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACxB,CAAC;AAED;;GAEG;AACH,SAAS,kBAAkB,CAAC,IAAY;IACtC,iFAAiF;IACjF,OAAO,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;AACzC,CAAC"}

package/dist/distill/stages/02-normalize.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Stage 2 — Normalize (deterministic)
+ *
+ * - Windows backslash paths → forward slash
+ * - Absolute paths → home-relative (~/)
+ * - ISO-8601 timestamp normalization (various formats → canonical)
+ * - Agent name casing → lowercase canonical
+ */
+export declare function normalize(text: string): string;

package/dist/distill/stages/02-normalize.js

@@ -0,0 +1,76 @@
+/**
+ * Stage 2 — Normalize (deterministic)
+ *
+ * - Windows backslash paths → forward slash
+ * - Absolute paths → home-relative (~/)
+ * - ISO-8601 timestamp normalization (various formats → canonical)
+ * - Agent name casing → lowercase canonical
+ */
+import { homedir } from 'node:os';
+/**
+ * Canonical agent name mapping. Keys are lowercase patterns to match,
+ * values are canonical forms.
+ */
+const AGENT_ALIASES = {
+    'claude-code': 'claude-code',
+    'claude code': 'claude-code',
+    'claudecode': 'claude-code',
+    'cursor': 'cursor',
+    'copilot': 'github-copilot',
+    'github copilot': 'github-copilot',
+    'github-copilot': 'github-copilot',
+};
+/** Regex for Windows-style absolute paths: C:\... or D:\... */
+const WIN_ABS_PATH_RE = /[A-Za-z]:\\(?:[^\s"'<>|?*\r\n]+)/g;
+/** Regex for Unix-style absolute paths starting with / */
+const UNIX_ABS_PATH_RE = /(?<!\w)(\/[^\s"'<>|?*\r\n:]{2,})/g;
+/**
+ * Common timestamp patterns to normalize to ISO-8601 UTC:
+ * - 2024-01-15T10:30:00.000Z  → keep as-is (already ISO)
+ * - 2024-01-15 10:30:00       → 2024-01-15T10:30:00Z
+ * - Jan 15, 2024 10:30 AM     → skip (too ambiguous without timezone)
+ */
+const DATETIME_SPACE_RE = /(\d{4}-\d{2}-\d{2})\s+(\d{2}:\d{2}:\d{2})(?:\.\d+)?(?:\s*UTC|\s*Z)?/g;
+const home = homedir();
+export function normalize(text) {
+    let result = text;
+    // 1. Windows paths → forward slash, then treat as absolute
+    result = result.replace(WIN_ABS_PATH_RE, (match) => {
+        // C:\Users\foo\bar → ~/bar (replace home dir portion)
+        const forwardSlash = match.replace(/\\/g, '/');
+        return homeRelative(forwardSlash);
+    });
+    // 2. Unix absolute paths → home-relative where applicable
+    result = result.replace(UNIX_ABS_PATH_RE, (match) => {
+        return homeRelative(match);
+    });
+    // 3. Datetime with space separator → ISO-8601
+    result = result.replace(DATETIME_SPACE_RE, (_match, date, time) => {
+        return `${date}T${time}Z`;
+    });
+    // 4. Agent names → lowercase canonical
+    result = normalizeAgentNames(result);
+    return result;
+}
+function homeRelative(path) {
+    // Normalize home dir to forward slashes for comparison
+    const homeForward = home.replace(/\\/g, '/');
+    const pathForward = path.replace(/\\/g, '/');
+    if (pathForward.startsWith(homeForward)) {
+        return '~' + pathForward.slice(homeForward.length);
+    }
+    // Also handle Windows-style home in forward-slash form: /Users/xxx or /home/xxx
+    // covered by the homedir() call above already
+    return pathForward;
+}
+function normalizeAgentNames(text) {
+    // Replace known agent name patterns (word-boundary aware)
+    for (const [alias, canonical] of Object.entries(AGENT_ALIASES)) {
+        // Escape special regex chars in alias
+        const escaped = alias.replace(/[-[\]{}()*+?.,\\^$|#\s]/g, '\\$&');
+        const re = new RegExp(`\\b${escaped}\\b`, 'gi');
+        text = text.replace(re, canonical);
+    }
+    return text;
+}
+//# sourceMappingURL=02-normalize.js.map