akm-cli 0.9.0-beta.53 → 0.9.0-beta.55

package/dist/storage/repositories/proposals-repository.js

@@ -0,0 +1,220 @@
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this
+// file, You can obtain one at https://mozilla.org/MPL/2.0/.
+/**
+ * Convert a raw `ProposalRow` to the public `Proposal` shape.
+ */
+export function proposalRowToProposal(row) {
+    let frontmatter;
+    if (row.frontmatter_json) {
+        try {
+            frontmatter = JSON.parse(row.frontmatter_json);
+        }
+        catch {
+            /* ignore corrupt frontmatter JSON */
+        }
+    }
+    let meta = {};
+    try {
+        meta = JSON.parse(row.metadata_json);
+    }
+    catch {
+        /* ignore */
+    }
+    return {
+        id: row.id,
+        ref: row.ref,
+        status: row.status,
+        source: row.source,
+        ...(typeof meta.sourceRun === "string" ? { sourceRun: meta.sourceRun } : {}),
+        createdAt: row.created_at,
+        updatedAt: row.updated_at,
+        payload: {
+            content: row.content,
+            ...(frontmatter !== undefined ? { frontmatter } : {}),
+        },
+        ...(meta.review !== undefined ? { review: meta.review } : {}),
+        ...(typeof meta.confidence === "number" ? { confidence: meta.confidence } : {}),
+        ...(meta.gateDecision !== undefined ? { gateDecision: meta.gateDecision } : {}),
+        ...(typeof meta.backupContent === "string" ? { backupContent: meta.backupContent } : {}),
+        ...(typeof meta.eligibilitySource === "string"
+            ? { eligibilitySource: meta.eligibilitySource }
+            : {}),
+    };
+}
+/**
+ * Convert a public `Proposal` to column values ready for an INSERT/UPDATE.
+ * The `stash_dir` comes from the call site (proposals.ts has it in scope).
+ */
+export function proposalToRowValues(proposal, stashDir) {
+    // Fields that have no dedicated column live in metadata_json.
+    const metaObj = {};
+    if (proposal.sourceRun !== undefined)
+        metaObj.sourceRun = proposal.sourceRun;
+    if (proposal.review !== undefined)
+        metaObj.review = proposal.review;
+    if (proposal.confidence !== undefined)
+        metaObj.confidence = proposal.confidence;
+    if (proposal.gateDecision !== undefined)
+        metaObj.gateDecision = proposal.gateDecision;
+    if (proposal.backupContent !== undefined)
+        metaObj.backupContent = proposal.backupContent;
+    if (proposal.eligibilitySource !== undefined)
+        metaObj.eligibilitySource = proposal.eligibilitySource;
+    return {
+        id: proposal.id,
+        stash_dir: stashDir,
+        ref: proposal.ref,
+        status: proposal.status,
+        source: proposal.source,
+        created_at: proposal.createdAt,
+        updated_at: proposal.updatedAt,
+        content: proposal.payload.content,
+        frontmatter_json: proposal.payload.frontmatter ? JSON.stringify(proposal.payload.frontmatter) : null,
+        metadata_json: JSON.stringify(metaObj),
+    };
+}
+/**
+ * Upsert a proposal row. Called by the proposal write path when state.db is
+ * the active backend.
+ */
+export function upsertProposal(db, proposal, stashDir) {
+    const v = proposalToRowValues(proposal, stashDir);
+    db.prepare(`
+    INSERT INTO proposals
+      (id, stash_dir, ref, status, source, created_at, updated_at, content, frontmatter_json, metadata_json)
+    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    ON CONFLICT(id) DO UPDATE SET
+      stash_dir        = excluded.stash_dir,
+      ref              = excluded.ref,
+      status           = excluded.status,
+      source           = excluded.source,
+      updated_at       = excluded.updated_at,
+      content          = excluded.content,
+      frontmatter_json = excluded.frontmatter_json,
+      metadata_json    = excluded.metadata_json
+  `).run(v.id, v.stash_dir, v.ref, v.status, v.source, v.created_at, v.updated_at, v.content, v.frontmatter_json, v.metadata_json);
+}
+/**
+ * List proposals, optionally filtered by stashDir, status, and/or ref.
+ *
+ * Results are ordered by `created_at ASC` (matching the historical
+ * `listProposals()` sort), with `rowid` as a deterministic tiebreak so two
+ * proposals created in the same millisecond list in insertion order.
+ */
+export function listStateProposals(db, options = {}) {
+    const conditions = [];
+    const params = [];
+    if (options.stashDir) {
+        conditions.push("stash_dir = ?");
+        params.push(options.stashDir);
+    }
+    if (options.status) {
+        conditions.push("status = ?");
+        params.push(options.status);
+    }
+    if (options.ref) {
+        conditions.push("ref = ?");
+        params.push(options.ref);
+    }
+    const where = conditions.length > 0 ? `WHERE ${conditions.join(" AND ")}` : "";
+    const rows = db
+        .prepare(`SELECT id, stash_dir, ref, status, source, created_at, updated_at,
+              content, frontmatter_json, metadata_json
+       FROM proposals ${where} ORDER BY created_at ASC, rowid ASC`)
+        .all(...params);
+    return rows.map(proposalRowToProposal);
+}
+/**
+ * Read every proposal's `gateDecision` record across all stashes (#612).
+ *
+ * Calibration reads the auto-accept gate's per-proposal decisions regardless of
+ * the proposal's current lifecycle status — a proposal that was auto-accepted
+ * is now `accepted`, an auto-rejected one stays `pending`, so filtering by
+ * status would drop half the join. Rows without a `gateDecision` (created
+ * before #577, or never gated) are skipped. The result is ordered by
+ * `decidedAt ASC` for deterministic downstream aggregation, falling back to
+ * `created_at` ordering from the SQL layer for rows with equal/missing
+ * timestamps.
+ */
+export function listProposalGateDecisions(db) {
+    const rows = db.prepare("SELECT metadata_json FROM proposals ORDER BY created_at ASC, rowid ASC").all();
+    const decisions = [];
+    for (const row of rows) {
+        let meta;
+        try {
+            meta = JSON.parse(row.metadata_json);
+        }
+        catch {
+            continue;
+        }
+        const decision = meta.gateDecision;
+        if (decision && typeof decision === "object" && typeof decision.outcome === "string") {
+            decisions.push(decision);
+        }
+    }
+    decisions.sort((a, b) => new Date(a.decidedAt).getTime() - new Date(b.decidedAt).getTime());
+    return decisions;
+}
+/**
+ * Look up a single proposal by id, optionally scoped to one stash root.
+ * Returns undefined when not found.
+ */
+export function getStateProposal(db, id, stashDir) {
+    const sql = `SELECT id, stash_dir, ref, status, source, created_at, updated_at,
+              content, frontmatter_json, metadata_json
+       FROM proposals WHERE id = ?${stashDir ? " AND stash_dir = ?" : ""}`;
+    const row = (stashDir ? db.prepare(sql).get(id, stashDir) : db.prepare(sql).get(id));
+    return row ? proposalRowToProposal(row) : undefined;
+}
+/**
+ * Find PENDING proposal ids in one stash whose id starts with `idPrefix`.
+ * Backs the UUID-prefix form of `akm proposal show/accept/... <prefix>` —
+ * prefix resolution is deliberately scoped to the live (pending) queue,
+ * mirroring the historical behaviour of scanning only the live directory.
+ *
+ * `%` / `_` / `\` in the prefix are escaped so the LIKE pattern is literal.
+ */
+export function listStateProposalIdsByPrefix(db, stashDir, idPrefix) {
+    const escaped = idPrefix.replace(/[\\%_]/g, (ch) => `\\${ch}`);
+    const rows = db
+        .prepare(`SELECT id FROM proposals
+       WHERE stash_dir = ? AND status = 'pending' AND id LIKE ? ESCAPE '\\'
+       ORDER BY id ASC`)
+        .all(stashDir, `${escaped}%`);
+    return rows.map((r) => r.id);
+}
+/**
+ * Whether the legacy filesystem proposal import has already run for `stashDir`.
+ * See migration 005 (`proposal_fs_imports`).
+ */
+export function hasImportedFsProposals(db, stashDir) {
+    // Drivers disagree on the no-row sentinel (bun:sqlite → null,
+    // better-sqlite3 → undefined) — Boolean() covers both.
+    return Boolean(db.prepare("SELECT 1 FROM proposal_fs_imports WHERE stash_dir = ?").get(stashDir));
+}
+/**
+ * Record that the legacy filesystem proposal import completed for `stashDir`
+ * so subsequent invocations skip the directory walk. INSERT OR REPLACE keeps
+ * the call idempotent.
+ */
+export function recordFsProposalsImport(db, stashDir, importedCount) {
+    db.prepare("INSERT OR REPLACE INTO proposal_fs_imports (stash_dir, imported_at, imported_count) VALUES (?, ?, ?)").run(stashDir, new Date().toISOString(), importedCount);
+}
+/**
+ * Insert a proposal row ONLY when the id is not already present (used by the
+ * legacy filesystem import so re-runs never clobber rows that have since been
+ * mutated through the canonical store). Returns true when a row was inserted.
+ */
+export function insertProposalIfAbsent(db, proposal, stashDir) {
+    const v = proposalToRowValues(proposal, stashDir);
+    const result = db
+        .prepare(`
+      INSERT OR IGNORE INTO proposals
+        (id, stash_dir, ref, status, source, created_at, updated_at, content, frontmatter_json, metadata_json)
+      VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    `)
+        .run(v.id, v.stash_dir, v.ref, v.status, v.source, v.created_at, v.updated_at, v.content, v.frontmatter_json, v.metadata_json);
+    const changes = result.changes ?? 0;
+    return Number(changes) > 0;
+}

package/dist/storage/repositories/recombine-repository.js

@@ -0,0 +1,213 @@
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this
+// file, You can obtain one at https://mozilla.org/MPL/2.0/.
+/**
+ * Record an induction of a recombine hypothesis and return the new consecutive
+ * count. INSERT … ON CONFLICT increments the streak, but the `last_run` guard
+ * makes a repeated call within the SAME run idempotent (no double-increment if
+ * the same ref appears twice in one run). On insert the streak starts at 1.
+ */
+export function recordRecombineInduction(db, input) {
+    const row = db
+        .prepare(`
+      INSERT INTO recombine_hypotheses
+        (hypothesis_ref, signature, member_key, consecutive_count, first_seen_at, last_seen_at, last_run)
+      VALUES (?, ?, ?, 1, ?, ?, ?)
+      ON CONFLICT(hypothesis_ref) DO UPDATE SET
+        consecutive_count = consecutive_count + (CASE WHEN last_run IS excluded.last_run THEN 0 ELSE 1 END),
+        last_seen_at = excluded.last_seen_at,
+        last_run = excluded.last_run,
+        signature = excluded.signature,
+        member_key = excluded.member_key
+      RETURNING consecutive_count
+    `)
+        .get(input.hypothesisRef, input.signature, input.memberKey, input.seenAt, input.seenAt, input.run);
+    return row?.consecutive_count ?? 0;
+}
+/**
+ * #633 — find an existing pending (non-promoted) hypothesis row whose cluster
+ * is the SAME generalization as a newly-induced one, matched by SIGNATURE plus
+ * a Jaccard membership-overlap test, rather than an exact member-set hash.
+ *
+ * In a growing stash any added/removed memory changes the exact member set, so
+ * the ref hash (and member_key) shift every run → a fresh row at count=1 → the
+ * streak never reaches `confirmThreshold` and nothing ever promotes. Matching
+ * on overlap lets a drifting-but-stable cluster keep accumulating under one row.
+ *
+ * Returns the matched row with the HIGHEST overlap (ties broken by most-recent
+ * `last_seen_at`), or `undefined` when none clears `minOverlap`. Already-promoted
+ * rows are ignored so a confirmed lesson is not reopened by a later induction.
+ *
+ * @param memberKey  the candidate cluster's membership fingerprint
+ *                   (sorted member entryKeys joined by `|`).
+ * @param minOverlap Jaccard threshold in [0,1]; a candidate matches when
+ *                   |A∩B| / |A∪B| >= minOverlap.
+ */
+export function findMatchingRecombineHypothesis(db, input) {
+    const candidateMembers = new Set(input.memberKey.split("|").filter((m) => m.length > 0));
+    if (candidateMembers.size === 0)
+        return undefined;
+    const rows = db
+        .prepare("SELECT * FROM recombine_hypotheses WHERE signature = ? AND promoted_at IS NULL ORDER BY last_seen_at DESC")
+        .all(input.signature);
+    let best;
+    let bestOverlap = -1;
+    for (const row of rows) {
+        const rowMembers = row.member_key.split("|").filter((m) => m.length > 0);
+        if (rowMembers.length === 0)
+            continue;
+        let intersection = 0;
+        for (const m of rowMembers) {
+            if (candidateMembers.has(m))
+                intersection += 1;
+        }
+        const union = candidateMembers.size + rowMembers.length - intersection;
+        const overlap = union === 0 ? 0 : intersection / union;
+        // rows are ordered last_seen_at DESC, so a strict `>` keeps the most-recent
+        // row on ties.
+        if (overlap >= input.minOverlap && overlap > bestOverlap) {
+            best = row;
+            bestOverlap = overlap;
+        }
+    }
+    return best;
+}
+/**
+ * Fetch a single recombine hypothesis row, or `undefined` when the ref has
+ * never been induced. Normalizes bun:sqlite null → undefined like
+ * {@link getExtractedSession}.
+ */
+export function getRecombineHypothesis(db, hypothesisRef) {
+    const row = db
+        .prepare("SELECT * FROM recombine_hypotheses WHERE hypothesis_ref = ?")
+        .get(hypothesisRef);
+    return row ?? undefined;
+}
+/**
+ * Mark a hypothesis promoted: stamp `promoted_at` and reset the consecutive
+ * count to 0, so it must re-accumulate a full confirmation streak before it can
+ * promote again. The `promoted_at` non-null state is the double-promotion guard.
+ */
+export function markRecombineHypothesisPromoted(db, hypothesisRef, promotedAt) {
+    db.prepare("UPDATE recombine_hypotheses SET promoted_at = ?, consecutive_count = 0 WHERE hypothesis_ref = ?").run(promotedAt, hypothesisRef);
+}
+/**
+ * #658 — does any current-run cluster match this hypothesis row under the SAME
+ * signature + Jaccard-overlap rule used for re-induction? A match means the
+ * cluster genuinely re-formed this run (it was merely cap-displaced out of the
+ * processed top-`maxClustersPerRun` slice), so its streak must NOT be reset.
+ */
+function hypothesisMatchesAnyPresentCluster(row, presentClusters, minOverlap) {
+    const rowMembers = row.member_key.split("|").filter((m) => m.length > 0);
+    if (rowMembers.length === 0)
+        return false;
+    const rowSet = new Set(rowMembers);
+    for (const cluster of presentClusters) {
+        if (cluster.signature !== row.signature)
+            continue;
+        const clusterMembers = cluster.memberKey.split("|").filter((m) => m.length > 0);
+        if (clusterMembers.length === 0)
+            continue;
+        let intersection = 0;
+        for (const m of clusterMembers) {
+            if (rowSet.has(m))
+                intersection += 1;
+        }
+        const union = rowSet.size + clusterMembers.length - intersection;
+        const overlap = union === 0 ? 0 : intersection / union;
+        if (overlap >= minOverlap)
+            return true;
+    }
+    return false;
+}
+/**
+ * Decay-to-zero every NON-promoted hypothesis NOT re-induced in the current run.
+ *
+ * A generalization that stops being supported by the corpus has lost its
+ * confirmation streak, so we hard-reset `consecutive_count` to 0 (the
+ * alternative — `count - 1` floored at 0 — tolerates a single noisy run but
+ * blurs the "consecutive" semantics; hard-reset is the conservative choice).
+ *
+ * Only rows whose `hypothesis_ref` is NOT in `seenRefs` AND whose `last_run` is
+ * NOT the current run are decayed. Already-promoted rows are left alone.
+ *
+ * #658 — CAP-AWARE decay. The recombine pass only re-inducts (and thus marks
+ * `seen`) the top-`maxClustersPerRun` clusters, but a cluster genuinely
+ * re-forms every run even when it is displaced below that cap. Resetting such a
+ * row treats a SCHEDULING miss as a SUBSTANCE miss and traps the hypothesis
+ * below `confirmThreshold` forever. When `opts.presentClusters` is supplied, a
+ * row is SPARED from decay if it Jaccard-matches any present cluster (same
+ * signature, overlap >= `opts.minOverlap`) — i.e. its cluster re-formed this run
+ * but was cap-displaced. This does NOT advance the streak (only re-induction in
+ * the processed slice does that, via {@link recordRecombineInduction}), so the
+ * recurrence bar for promotion is unchanged; it only stops the cap from
+ * manufacturing artificial misses. Omitting `presentClusters` preserves the
+ * pre-#658 hard-reset-after-one-miss behaviour exactly.
+ *
+ * Returns the number of rows reset.
+ */
+export function decayUnseenRecombineHypotheses(db, currentRun, seenRefs, opts) {
+    // #658 — when cap-aware sparing is requested, fold the cap-displaced rows into
+    // the "seen" exclusion set: the underlying reset SQL already protects every
+    // ref it is given, so sparing == treating a spared row exactly like a seen
+    // row for this sweep (its count is left untouched, never advanced).
+    let effectiveSeen = seenRefs;
+    if (opts && opts.presentClusters.length > 0) {
+        const candidates = db
+            .prepare("SELECT hypothesis_ref, signature, member_key FROM recombine_hypotheses WHERE promoted_at IS NULL AND (last_run IS NULL OR last_run != ?) AND consecutive_count != 0")
+            .all(currentRun);
+        const seenSet = new Set(seenRefs);
+        for (const row of candidates) {
+            if (seenSet.has(row.hypothesis_ref))
+                continue;
+            if (hypothesisMatchesAnyPresentCluster(row, opts.presentClusters, opts.minOverlap)) {
+                seenSet.add(row.hypothesis_ref);
+            }
+        }
+        effectiveSeen = [...seenSet];
+    }
+    return decayUnseenRecombineHypothesesInner(db, currentRun, effectiveSeen);
+}
+/**
+ * The raw reset sweep shared by the cap-aware wrapper above. Resets every
+ * non-promoted row from a prior run whose ref is NOT in `seenRefs`. Kept private
+ * so the param-ceiling chunking logic lives in one place.
+ */
+function decayUnseenRecombineHypothesesInner(db, currentRun, seenRefs) {
+    // Reset every eligible row, then exclude the seen refs in chunks to respect
+    // the ~999 SQLite param ceiling. With no seen refs we reset all non-promoted
+    // rows from prior runs in a single statement.
+    if (seenRefs.length === 0) {
+        const res = db
+            .prepare("UPDATE recombine_hypotheses SET consecutive_count = 0 WHERE promoted_at IS NULL AND (last_run IS NULL OR last_run != ?) AND consecutive_count != 0")
+            .run(currentRun);
+        return Number(res.changes);
+    }
+    // A single NOT IN keeps the exclusion atomic (a chunked NOT IN would let a ref
+    // excluded by one chunk still be reset by another chunk's statement). The
+    // recombine pass caps RE-INDUCED clusters at `maxClustersPerRun` (a handful) —
+    // but with #658 cap-aware sparing the caller folds every cap-displaced
+    // (present-but-unprocessed) hypothesis into `effectiveSeen` too, so on a large
+    // stash `seenRefs` here can carry MANY spared refs, not just the handful that
+    // were processed. We cap defensively at ~900 (under SQLite's ~999 param
+    // ceiling): if `effectiveSeen` somehow exceeds it we fall back to resetting all
+    // eligible rows — which re-introduces the cap-displacement trap for THAT run
+    // (spared rows get decayed because the NOT IN protection is dropped). That is a
+    // rare, bounded degradation; a stash with >900 simultaneously-spared
+    // hypotheses is far beyond current scale.
+    if (seenRefs.length > 900) {
+        const res = db
+            .prepare("UPDATE recombine_hypotheses SET consecutive_count = 0 WHERE promoted_at IS NULL AND (last_run IS NULL OR last_run != ?) AND consecutive_count != 0")
+            .run(currentRun);
+        return Number(res.changes);
+    }
+    const placeholders = seenRefs.map(() => "?").join(",");
+    const res = db
+        .prepare(`UPDATE recombine_hypotheses SET consecutive_count = 0
+       WHERE promoted_at IS NULL
+         AND (last_run IS NULL OR last_run != ?)
+         AND consecutive_count != 0
+         AND hypothesis_ref NOT IN (${placeholders})`)
+        .run(currentRun, ...seenRefs);
+    return Number(res.changes);
+}

package/dist/storage/repositories/task-history-repository.js

@@ -0,0 +1,93 @@
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this
+// file, You can obtain one at https://mozilla.org/MPL/2.0/.
+/**
+ * Upsert a task history row.
+ */
+export function upsertTaskHistory(db, row) {
+    // INSERT OR IGNORE: if a run with the same (task_id, started_at) was already
+    // imported (e.g. by the migration script), skip it silently.
+    db.prepare(`
+    INSERT OR IGNORE INTO task_history
+      (task_id, status, started_at, completed_at, failed_at, log_path,
+       target_kind, target_ref, metadata_json)
+    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+  `).run(row.task_id, row.status, row.started_at, row.completed_at ?? null, row.failed_at ?? null, row.log_path ?? null, row.target_kind ?? null, row.target_ref ?? null, row.metadata_json);
+}
+/**
+ * Look up a task history row by task_id. Returns undefined when not found.
+ */
+/**
+ * Return the most recent run for a given task_id, or undefined if no runs exist.
+ */
+export function getTaskHistory(db, taskId) {
+    return db
+        .prepare(`SELECT id, task_id, status, started_at, completed_at, failed_at, log_path,
+              target_kind, target_ref, metadata_json
+       FROM task_history WHERE task_id = ? ORDER BY started_at DESC LIMIT 1`)
+        .get(taskId);
+}
+/**
+ * Return all runs for a given task_id, newest first.
+ */
+export function getTaskHistoryRuns(db, taskId, limit = 50) {
+    return db
+        .prepare(`SELECT id, task_id, status, started_at, completed_at, failed_at, log_path,
+              target_kind, target_ref, metadata_json
+       FROM task_history WHERE task_id = ? ORDER BY started_at DESC LIMIT ?`)
+        .all(taskId, limit);
+}
+/**
+ * Query task history rows by started_at range and/or status.
+ */
+export function queryTaskHistory(db, options = {}) {
+    const conditions = [];
+    const params = [];
+    if (options.since) {
+        conditions.push("started_at >= ?");
+        params.push(options.since);
+    }
+    if (options.until) {
+        conditions.push("started_at <= ?");
+        params.push(options.until);
+    }
+    if (options.status) {
+        conditions.push("status = ?");
+        params.push(options.status);
+    }
+    if (options.targetKind) {
+        conditions.push("target_kind = ?");
+        params.push(options.targetKind);
+    }
+    if (options.targetRef) {
+        conditions.push("target_ref = ?");
+        params.push(options.targetRef);
+    }
+    const where = conditions.length > 0 ? `WHERE ${conditions.join(" AND ")}` : "";
+    return db
+        .prepare(`SELECT task_id, status, started_at, completed_at, failed_at, log_path,
+              target_kind, target_ref, metadata_json
+       FROM task_history ${where} ORDER BY started_at DESC`)
+        .all(...params);
+}
+/**
+ * Read COMPLETED `akm-improve` task_history runs whose `started_at` falls in
+ * `[since, until)` (or `started_at >= since` when `until` is omitted), ordered
+ * oldest-first by `started_at`. Only rows with a non-null `completed_at` are
+ * returned (in-flight runs are excluded). The `task_id = 'akm-improve'`
+ * predicate is fixed because the only caller (commands/health.ts
+ * `loadTaskIntervals`) builds wall-time intervals for the improve cron task.
+ *
+ * Owns the SQL formerly inlined in commands/health.ts. Note the bound is
+ * EXCLUSIVE on the upper end (`started_at < ?`) — callers pass an already
+ * widened window; this helper does not widen.
+ *
+ * Connection-lifetime rule (WS5): `.all()` materializes a plain array before
+ * returning.
+ */
+export function queryCompletedTaskIntervals(db, since, until) {
+    const sql = until
+        ? "SELECT started_at, completed_at FROM task_history WHERE task_id = 'akm-improve' AND started_at >= ? AND started_at < ? AND completed_at IS NOT NULL ORDER BY started_at"
+        : "SELECT started_at, completed_at FROM task_history WHERE task_id = 'akm-improve' AND started_at >= ? AND completed_at IS NOT NULL ORDER BY started_at";
+    return (until ? db.prepare(sql).all(since, until) : db.prepare(sql).all(since));
+}

package/dist/storage/sqlite-pragmas.js

@@ -57,8 +57,8 @@ export function resolveJournalMode(raw) {
  * `process.env.AKM_SQLITE_JOURNAL_MODE`. Read at call time (per open) so tests
  * that set the env per-case see the right value and we avoid stale-env flakes.
  */
-export function resolveConfiguredJournalMode() {
-    return resolveJournalMode(process.env.AKM_SQLITE_JOURNAL_MODE);
+export function resolveConfiguredJournalMode(env = process.env) {
+    return resolveJournalMode(env.AKM_SQLITE_JOURNAL_MODE);
 }
 function warnInvalidJournalModeOnce(raw) {
     if (warnedInvalid)
@@ -111,7 +111,7 @@ export function isNetworkFilesystem(fsType) {
  * WAL path emits no `synchronous` pragma, exactly as before.
  */
 export function applyStandardPragmas(db, opts = {}) {
-    let mode = resolveConfiguredJournalMode();
+    let mode = resolveConfiguredJournalMode(opts.env);
     // Network-FS fallback only fires for the WAL default and only when we have a
     // directory to probe. An explicitly-requested DELETE/TRUNCATE is never
     // overridden, and a failed/unsupported probe (undefined) keeps WAL.

package/dist/tasks/backends/index.js

@@ -4,7 +4,14 @@
 import { CRON_BACKEND } from "./cron.js";
 import { LAUNCHD_BACKEND } from "./launchd.js";
 import { SCHTASKS_BACKEND } from "./schtasks.js";
+let backendsOverrides;
+/** TEST-ONLY. Swap backend selection; pass undefined to restore the real implementations. */
+export function _setBackendsForTests(fakes) {
+    backendsOverrides = fakes;
+}
 export function selectBackend(options = {}) {
+    if (backendsOverrides?.selectBackend)
+        return backendsOverrides.selectBackend(options);
     const platform = options.platform ?? process.platform;
     switch (platform) {
         case "win32":
@@ -16,6 +23,8 @@ export function selectBackend(options = {}) {
     }
 }
 export function backendNameForPlatform(platform = process.platform) {
+    if (backendsOverrides?.backendNameForPlatform)
+        return backendsOverrides.backendNameForPlatform(platform);
     if (platform === "win32")
         return "schtasks";
     if (platform === "darwin")

package/dist/tasks/runner.js

@@ -33,13 +33,14 @@ import { loadConfig } from "../core/config/config.js";
 import { NotFoundError, rethrowIfTestIsolationError } from "../core/errors.js";
 import { buildTaskRunId, insertTaskLogLines, openLogsDatabase, } from "../core/logs-db.js";
 import { getTaskLogDir } from "../core/paths.js";
-import { getTaskHistory, queryTaskHistory, upsertTaskHistory, withStateDb } from "../core/state-db.js";
+import { withStateDb } from "../core/state-db.js";
 import { error } from "../core/warn.js";
 import { requireAgentProfile, runAgent } from "../integrations/agent/index.js";
 import { resolveProcessAgentProfile } from "../integrations/agent/config.js";
 import { resolveRunner } from "../integrations/agent/runner.js";
 import { spawn } from "../runtime.js";
 import { resolveAssetPath } from "../sources/resolve.js";
+import { getTaskHistory, queryTaskHistory, upsertTaskHistory } from "../storage/repositories/task-history-repository.js";
 import { startWorkflowRun } from "../workflows/runtime/runs.js";
 import { parseTaskDocument } from "./parser.js";
 export async function runTask(id, options = {}) {
@@ -131,6 +132,11 @@ async function runCommandTask(input) {
             stdout: "pipe",
             stderr: "pipe",
             cwd: process.env.HOME ?? "/tmp",
+            // Stamp task-runner provenance so any akm invocation in the command tree
+            // records usage events as machine traffic, not user demand (DRIFT-6).
+            // A more specific stamp already in the environment (e.g. improve's
+            // AKM_EVENT_SOURCE=improve on its child spawns) still wins in children.
+            env: { ...process.env, AKM_EVENT_SOURCE: process.env.AKM_EVENT_SOURCE ?? "task" },
         });
         let timer;
         let timedOut = false;
@@ -351,6 +357,10 @@ async function runPromptTask(input) {
         timeoutMs: agentTimeoutMs,
         cwd: stashDir,
         ...agentOptions,
+        // Stamp task-runner provenance for any akm invocation the agent makes
+        // (DRIFT-6: agent-task traffic must not be recorded as user demand).
+        // Caller-supplied env still wins on conflicts.
+        env: { AKM_EVENT_SOURCE: "task", ...agentOptions?.env },
     });
     const finishedAt = now();
     const log = renderPromptLog({ task, profileName: profile.name, result });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akm-cli",
-  "version": "0.9.0-beta.53",
+  "version": "0.9.0-beta.55",
   "type": "module",
   "description": "akm (Agent Knowledge Management) — A package manager for AI agent skills, commands, tools, and knowledge. Works with Claude Code, OpenCode, Cursor, and any AI coding assistant.",
   "keywords": [
@@ -58,7 +58,7 @@
     "sweep:tmp": "bun scripts/sweep-test-tmp.ts",
     "test": "bash scripts/test-unit.sh",
     "test:unit": "bash scripts/test-unit.sh",
-    "test:integration": "bun run sweep:tmp && bun test --isolate --timeout=30000 ./tests/integration",
+    "test:integration": "bun run sweep:tmp && bun test --timeout=30000 ./tests/integration",
     "test:node-smoke": "bun scripts/node-smoke.ts",
     "test:node-compat": "AKM_NODE_COMPAT_TESTS=1 bun test --timeout=120000 tests/integration/node-compat.test.ts",
     "test:time": "bun scripts/test-timing-report.ts",