akm-cli 0.9.28-alpha.2 → 0.9.28-alpha.3

package/CHANGELOG.md

@@ -6,6 +6,14 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.9.28-alpha.3] - 2026-10-08
+
+### Changed
+
+- A consolidate pair-pass retire proposal now records the judge's claim lists in its `retirement` metadata as `onlyInRetired` and `onlyInSuccessor`, named by role rather than by the judge's A/B, so what the judge found only on each side survives past the run. Proposals minted before this lack both fields; no row is added for a pair akm keeps.
+
+- A distill lesson the quality gate rejects or sends to review now keeps its text: the `distill_invoked` event and the distill result carry `rejectedContent`, cut to 2000 characters (the judge prompt's cap). Before, a rejected lesson left only its score and reason. It stays local in `state.db`.
+
 ## [0.9.28-alpha.2] - 2026-10-08
 
 ### Added

package/dist/commands/improve/consolidate/pair-pass.js

@@ -637,6 +637,9 @@ async function judgeOne(ctx, candidate) {
         cosine: candidate.cosine,
         judgeLabel: verdict.relation,
         judgeReason: verdict.reason,
+        // The judge's A is the older side and B the newer (see orderByAge); record the lists by role.
+        onlyInRetired: retired === older ? verdict.onlyInA : verdict.onlyInB,
+        onlyInSuccessor: retired === older ? verdict.onlyInB : verdict.onlyInA,
         retiredContentHash: retiredHash,
         successorContentHash: successorHash,
         reason,

package/dist/commands/improve/distill.js

@@ -638,6 +638,8 @@ function rejectDistilled(run, proposalRef, content, score, reason, meta) {
         ledgerRef: run.ledgerRef,
     });
 }
+/** Longest lesson text kept on a rejection, the same cap the quality judge's prompt reads (stage.ts). */
+const REJECTED_CONTENT_MAX_CHARS = 2000;
 /**
  * Record a distill quality-gate outcome and return its envelope.
  * `quality_rejected` lands in the improve ledger under the input's key (its
@@ -673,6 +675,8 @@ export function writeQualityRejection(args) {
         }
     }
     const eligMeta = args.eligibilitySource ? { eligibilitySource: args.eligibilitySource } : {};
+    // The text the gate turned away, kept in the local event and the result so a rejection can be audited.
+    const rejectedContent = args.content.slice(0, REJECTED_CONTENT_MAX_CHARS);
     appendEvent({
         eventType: "distill_invoked",
         ref: ledgerRef,
@@ -681,6 +685,7 @@ export function writeQualityRejection(args) {
             proposalRef: args.proposalRef,
             score: args.score,
             reason: args.reason,
+            rejectedContent,
             ...meta,
             ...eligMeta,
         },
@@ -693,6 +698,7 @@ export function writeQualityRejection(args) {
         proposalRef: args.proposalRef,
         score: args.score,
         reason: args.reason,
+        rejectedContent,
         ...(proposal ? { proposalId: proposal.id, proposal } : {}),
         ...meta,
     };

package/docs/reference/data-and-telemetry.md

@@ -196,7 +196,7 @@ the set of types the code actually emits at HEAD (verified against every
 | `reflect_completed` | Reflect phase produced a proposal | `ref` |
 | `improve_reflect_outcome` | Per-asset reflect result | `ref`, `ok`, `durationMs`, `reason` |
 | `propose_invoked` | `akm proposal new` | `ref` |
-| `distill_invoked` | Distill phase inside the `akm improve`/`akm proposal new` pipeline. **`akm distill` is not a CLI command** — there is no standalone verb by that name | `ref`, outcome (`queued`, `skipped` with a `skipReason` such as `lesson_exists`, `nothing_reusable` or `conflict_noop`, `llm_failed`, `validation_failed`, `quality_rejected`, `review_needed`) |
+| `distill_invoked` | Distill phase inside the `akm improve`/`akm proposal new` pipeline. **`akm distill` is not a CLI command** — there is no standalone verb by that name | `ref`, outcome (`queued`, `skipped` with a `skipReason` such as `lesson_exists`, `nothing_reusable` or `conflict_noop`, `llm_failed`, `validation_failed`, `quality_rejected`, `review_needed`). A `quality_rejected` or `review_needed` event also carries `score`, `reason`, the judge's per-criterion `criteria` when it ran, and `rejectedContent`, the turned-away lesson text cut to 2000 characters (the judge prompt's cap), kept locally in `state.db` |
 | `extract_invoked` | `akm proposal extract --type <harness>` / `--auto`, or improve-stage session extraction | `outcome`, `sessionId`, `harness` |
 | `extract_triaged` | The pre-LLM extract triage gate evaluated at least one session | `evaluated`, `passed`, `triagedOut`, `sourceRun` (aggregated) |
 | `schema_repair_invoked` | The schema-repair pass inside `akm improve` (`runSchemaRepairPass`) attempts to patch missing frontmatter on an asset that failed schema validation. **There is no `akm lint --repair` flag** — `lint` has `--fix`/`--auto-fix`, unrelated to this event | `ref`, outcome |
@@ -313,6 +313,7 @@ Contents:
 - Source (which process generated it — e.g. `reflect`, `distill`)
 - Full proposal content (Markdown text)
 - Created/updated timestamps
+- For a `consolidate-pair` retire proposal, the pair judge's verdict in its `retirement` metadata: label, reason, cosine, and the claims only the retired side holds (`onlyInRetired`) and only the kept side holds (`onlyInSuccessor`), each at most 20 entries of 200 characters
 
 Beside it, the `improve_ledger` table records what each improve stage last did
 with each asset — one row per bundle, asset ref and stage: the outcome

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akm-cli",
-  "version": "0.9.28-alpha.2",
+  "version": "0.9.28-alpha.3",
   "type": "module",
   "description": "akm (Agent Knowledge Manager) — a portable, local-first capability library for AI agents. Discover, load, share, and improve reusable skills, scripts, workflows, and knowledge across any shell-capable coding agent, including Claude Code, OpenCode, and Cursor.",
   "keywords": [