akm-cli 0.9.26 → 0.9.27-alpha.2

package/docs/integration/bundling-akm.md

@@ -266,7 +266,7 @@ rather than rely on `$HOME`-derived defaults (names verified against
 | `AKM_CONFIG_DIR` | `config.json`'s directory. |
 | `AKM_DATA_DIR` | Durable, non-regenerable data: **`index.db` and `state.db` live here.** This is the directory a migration snapshot's safety copy sits beside. |
 | `AKM_CACHE_DIR` | Regenerable cache: registry downloads, config backups, task logs. Safe to discard between image builds (not between boots of the same running install). |
-| `AKM_STATE_DIR` | **Not** where `state.db` lives, despite the name — this is the XDG "state" directory. Holds scheduled-task invocation context, companion-plugin hook state (Claude Code / OpenCode hook logs), and, per stash, `akm improve`'s machine-local writers (`improve/measurement/verdicts/`) and whole-run lock (`locks/`) — see [Storage locations](https://github.com/itlackey/akm/blob/main/docs/architecture/internals/storage-locations.md). Set it anyway if you schedule akm tasks inside the image, so that context is captured consistently rather than falling back to `$HOME/.local/state/akm`. |
+| `AKM_STATE_DIR` | **Not** where `state.db` lives, despite the name — this is the XDG "state" directory. Holds scheduled-task invocation context, companion-plugin hook state (Claude Code / OpenCode hook logs), and, per stash, `akm improve`'s whole-run lock (`locks/`) — see [Storage locations](https://github.com/itlackey/akm/blob/main/docs/architecture/internals/storage-locations.md). Set it anyway if you schedule akm tasks inside the image, so that context is captured consistently rather than falling back to `$HOME/.local/state/akm`. |
 
 Set all five to paths that persist across container restarts (a mounted
 volume), or `akm migrate apply` will see an empty `state.db` on every boot

package/docs/migration/v0.8-to-v0.9.md

@@ -144,7 +144,9 @@ step is idempotent — a second run reports nothing pending. 0.9.17-alpha.4
 removed that step: `akm migrate` no longer relocates these files, and one left
 at an old path is inert (nothing reads it). Two of the five writers no longer
 exist either — the improve ledger replaced `distill-rejected/`, and the
-write-only `eval-cases/` path was removed.
+write-only `eval-cases/` path was removed. A third, `measurement/verdicts/`, went
+with the `scripts/akm-eval` toolkit that wrote it, which has since left this
+repository ([akm-eval](https://github.com/itlackey/akm/blob/main/docs/maintainers/eval.md)).
 `$STASH/.akm/memory-cleanup/` did not move; it is the one confirmed exception
 to the rule (see Storage locations, above).
 

package/docs/reference/README.md

@@ -17,4 +17,4 @@ Authoritative reference documentation for the akm CLI and its data.
 - [Website Sources](https://github.com/itlackey/akm/blob/main/docs/reference/website-sources.md) -- The pluggable fetcher API behind `akm import <url>` and other URL-based knowledge reads
 - [Data & Telemetry](data-and-telemetry.md) -- Exactly what akm reads and writes on your machine (no remote telemetry)
 
-See also: [akm-eval](https://github.com/itlackey/akm/blob/main/docs/maintainers/eval.md) -- the standalone toolkit for measuring whether `akm improve` is working (maintainer docs), and the repo-root [Roadmap](https://github.com/itlackey/akm/blob/main/ROADMAP.md) -- high-level focus for upcoming releases.
+See also: [akm-eval](https://github.com/itlackey/akm-eval) -- the evals and benchmarks for measuring whether `akm improve` is working, and the repo-root [Roadmap](https://github.com/itlackey/akm/blob/main/ROADMAP.md) -- high-level focus for upcoming releases.

package/docs/reference/cli.md

@@ -2492,16 +2492,19 @@ day; an asset a stage looked at and left unchanged is revisited after 7 days,
 or as soon as new feedback (or, for consolidation, an edit) arrives.
 
 Consolidation's promotion of a memory into `knowledge/` is the exception to the
-7-day rule: once a promotion is accepted or rejected, its memory is not offered
-to the model again until its body changes, however long that takes. The ledger
+7-day rule: once a promotion is accepted or rejected, or the model judged the
+memory and proposed nothing, the memory is not offered to the model again until
+its body changes, however long that takes. The ledger
 records the body hash the promotion was decided against and compares it with
 the memory's current body (frontmatter edits do not count), the same
 content-driven rule the consolidate pair pass uses. A promotion decided by an
-older release, which recorded no hash, keeps the old windows. Consolidation
+older release, which recorded no hash, keeps the old windows. A memory whose
+body equals that of a consolidate promotion rejected on or after 2026-09-29 is
+held the same way, under whatever name it has. Consolidation
 also does not promote a memory that `knowledge/` already covers: before it
 queues a promotion it compares the memory with the 20 `knowledge/` docs in its
 bundle nearest to it by stored vector, and skips the memory when one of them
-holds at least half of its distinct 5-word shingles (skip reason
+holds at least 30% of its distinct 5-word shingles (skip reason
 `dedup_covered_by_knowledge` in the result's `consolidation.skipReasons`). A
 covering doc that ranks lower than the 20th nearest goes unseen. With no stored
 vector (semantic search off, or the memory not indexed yet) that check does
@@ -2780,6 +2783,7 @@ harvest" branch on `skipReasons`, `warnings`, or `sessionsProcessed` /
 | `engineKind` | `"llm"`, `"sdk"`, or `"agent"` — the kind of runner `engine` resolved to. Same absence condition as `engine`. |
 | `skipReasons` | Per-`skipReason` count across `sessions[]` (e.g. `{ "llm_unavailable": 25 }`). Present only when `sessionsSkipped > 0`. |
 | `warnings` | Includes one aggregate line per infrastructure skip reason that fired (`llm_unavailable`, `read_failed`, `exception`, `locked_concurrent`) — e.g. `25 of 25 sessions skipped: llm_unavailable (engine "default")` — so an engine outage is visible without inspecting `sessions[]`. Session-content skips (`already_extracted`, `too_short`, `triaged_out`) are counted in `skipReasons` but never produce a warning line. |
+| `sessions[].warnings` | One line per candidate the model wrote that the output contract refuses, as `<type>:<name> dropped: <reason>` — a lesson without a `when_to_use` of 15 characters, a description under 20, a name that is not a kebab-case slug — and one per candidate held back by the improve ledger. A session whose every candidate was dropped has `candidateCount: 0` and no `rationaleIfEmpty`, so this is where the loss shows. |
 
 #### proposal new
 

package/docs/reference/configuration.md

@@ -110,8 +110,11 @@ distill quality-gate judges do too unless the judge's engine sets
 `reasoning_effort` when set. Backend support: llama.cpp direct honors both forms
 (`reasoning_effort` from build ≥ b10644); vLLM honors
 `chat_template_kwargs`; Bifrost drops `chat_template_kwargs` and passes
-`reasoning_effort` through, so also set `reasoningEffort: "none"` behind it; a
-strict hosted API may 400 on unrecognized keys. Both fields are AKM-owned, not
+`reasoning_effort` through, so also set `reasoningEffort: "none"` behind it. A
+strict hosted API that rejects the two fields (OpenAI answers 400 `Unknown
+parameter: 'chat_template_kwargs'`) gets one retry without both, and AKM stops
+sending them to that endpoint and model for the rest of the process, as it
+does for `response_format` below. Both fields are AKM-owned, not
 settable via `extraParams`. A response with reasoning tokens despite
 `enableThinking: false` triggers a runtime warning and the `akm health`
 `thinking-control` advisory.

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

@@ -195,7 +195,7 @@ 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` 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`) |
 | `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 |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akm-cli",
-  "version": "0.9.26",
+  "version": "0.9.27-alpha.2",
   "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": [