@event4u/agent-config 16.5.0 → 17.0.0

package/dist/agent-src/commands/fix/ci.md

@@ -25,16 +25,18 @@ A run still in progress has no verdict to fix. Do NOT read a partial run and
 start guessing — and do NOT end the turn reporting "CI is running".
 
 ```bash
-until [ "$(gh pr checks <n> --json state \
-    --jq '[.[] | select(.state=="IN_PROGRESS" or .state=="QUEUED" or .state=="PENDING")] | length')" = "0" ]
-do sleep 45; done
+npm exec --no -- agent-config ci:settle <n>
 ```
 
-Run that in the BACKGROUND so the turn resumes when the verdict lands; a
-foreground sleep-poll chain is blocked and per-minute re-polling is the tool
-loop `token-efficiency` forbids. If the waiter exits because `gh` itself failed
-(network down), it has observed nothing — say so rather than treating it as a
-pass.
+`npm exec --no --` reaches a project-local install and a global one alike and
+never fetches from the registry. `ci_settle` is the one waiter for this condition — never a hand-written
+`gh pr checks` loop. Run it in the BACKGROUND so the turn resumes when the
+verdict lands, and do not pipe it: its LAST output line is the verdict and its
+exit code carries it. Exit 0 is settled green for the named head; exit 1 is
+settled with at least one failed check — go to step 1; exit 2 is NOT a verdict
+(did not settle within its budget, `gh` could not be read, or the PR is merged,
+closed or head-diverged) — it has observed nothing, so say so and re-run or
+report, never treat it as a pass.
 
 Note `gh run view --log-failed` returns *"run is still in progress; logs will be
 available when it is complete"* until the WHOLE run finishes, which is the other

package/dist/agent-src/commands/ghostwriter/write.md

@@ -103,6 +103,28 @@ Generate the body as a single fenced markdown block per
 fingerprint constraints: sentence-length ±25 %, opener / closer
 patterns, hashtag / emoji rules, ±15 % length tolerance.
 
+### 5b. Check the finished draft
+
+Step 4 steers drafting; this step checks the result. Pipe the step-5
+block on stdin — this command never writes the draft to disk; the checker
+scans inside a single wrapping fence, any other fenced text stays exempt:
+
+```bash
+npx tsx node_modules/@event4u/agent-config/src/scripts/check_voice_lexicon.ts --draft - \
+  --ghostwriter agents/reference/ghostwriter/<slug>.md \
+  [--slot agents/context-spine/brand-voice.md]
+```
+
+Pass `--slot` only when `agent-config brand:status --json` reports the
+`brand-voice` slot valid. Exit **1** is a profile taboo in the draft:
+print its located finding and **abort** — no draft, no footer (§ Rules).
+Brand `avoid` hits and `unbacked-claim` findings flag without blocking:
+print each after the fenced block with its line, its note or the missing
+proof. Quoted, cited and code-fenced text is never scanned. Exit **2** is a
+broken invocation (unreadable profile, invalid slot, no input): say so and
+never present the draft as checked. No runtime → say so in one line and
+rely on step 4.
+
 ### 6. Append the disclosure footer (literal, deterministic)
 
 After the body, emit a blank line, then the literal string (rendered

package/dist/agent-src/commands/memory/learn-low-impact.md

@@ -5,9 +5,8 @@ pack: memory
 visibility: internal
 cluster: memory
 sub: learn-low-impact
-skills: [ai-council, upstream-contribute]
-description: Preview validated low-impact entries that would be upstreamed to the package seed (default `--preview`); `--apply` opens a draft PR via `upstream-contribute` after re-redaction.
-argument-hint: "[--preview|--apply]"
+skills: [ai-council]
+description: Preview which validated low-impact entries would be worth upstreaming — read-only, opens nothing, writes nothing.
 suggestion:
   eligible: false
   rationale: "Cluster sub-command — reached via its cluster head's routing or its explicit /cluster:sub name; not independently suggested (surface-consolidation)."
@@ -19,28 +18,30 @@ packs:
 
 # /memory learn-low-impact
 
-Promote `## Validated` entries from
+Show which `## Validated` entries from
 [`agents/decisions/low-impact-decisions.md`](../../agents/decisions/low-impact-decisions.md)
-into the upstream seed at
-`.agent-src.uncondensed/data/low-impact-decisions-seed.md` via a DRAFT
-PR against the agent-config package. **Validated entries only** — probation
-entries never upstream, they're unconfirmed signal.
+would be worth upstreaming, and which the privacy floor refuses.
+**Validated entries only** — probation entries are unconfirmed signal and
+never appear here.
 
-## Flags
+## This command opens nothing
 
-| Flag | Default | Behaviour |
-|---|---|---|
-| `--preview` | **on** | Build the plan, run the redactor, render promoted / refused / already-seeded buckets + draft PR body. **No file write, no branch, no PR.** Default behaviour. |
-| `--apply` | off | Mutually exclusive with `--preview`. Required to invoke `upstream-contribute` and open the draft PR. Refusals from the redactor still block. |
+```
+READ-ONLY. NO BRANCH, NO COMMIT, NO PUSH, NO PULL REQUEST, NO FILE WRITE.
+THE `--apply` FLAG WAS RETIRED: IT WROTE INTO A PACKAGE SEED FILE
+THAT DOES NOT EXIST, AND A SECOND WAY OUT WOULD BE A SECOND SET
+OF GUARANTEES. ONE WAY OUT IS `self-repair:release`.
+```
 
-Iron Law: ``--apply`` never auto-fires on the first invocation. The
-user always sees the preview block first and re-runs explicitly.
+The package ships no seed corpus today, so the already-seeded baseline is
+empty and every validated entry that clears the redactor shows as promotable.
+That is a true reading of the tree, not a backlog.
 
 ## Iron Law — privacy floor runs TWICE
 
 ```
-THE REDACTOR RUNS AT INTAKE (WRITE GATE) AND AGAIN HERE (UPSTREAM GATE).
-ANY VIOLATION → REFUSE THE PR. NO SILENT REWRITES.
+THE REDACTOR RUNS AT INTAKE (WRITE GATE) AND AGAIN HERE (REVIEW GATE).
+ANY VIOLATION → THE ENTRY IS REFUSED AND NAMED. NO SILENT REWRITES.
 ```
 
 See [`low-impact-corpus-privacy-floor`](../rules/low-impact-corpus-privacy-floor.md)
@@ -50,37 +51,55 @@ for the eight forbidden-content classes.
 
 ### 1. Build the preview plan
 
-Call
-`scripts/ai_council/learn_low_impact_preview.ts::build_preview` with
-the project-local corpus and the package seed:
-
-```python
-from scripts.ai_council.learn_low_impact_preview import build_preview
-plan = build_preview(
-    corpus_path="agents/decisions/low-impact-decisions.md",
-    seed_path=".agent-src.uncondensed/data/low-impact-decisions-seed.md",
-    repo_slug="<owner>/<repo>",  # from `git remote get-url origin`
-    repo_root="<absolute repo root>",
-    private_domains=(),   # from .agent-settings.yml policy
-    customer_names=(),    # from .agent-settings.yml policy
-    sql_identifiers=(),   # from .agent-settings.yml policy
-)
+Call `build_preview` from
+`src/scripts/ai_council/learn_low_impact_preview.ts` with the project-local
+corpus and the exported seed path:
+
+```ts
+import * as path from 'node:path';
+
+// Resolved against the package root — a bare `src/…` specifier does not resolve.
+const { build_preview, PACKAGE_SEED_PATH } = await import(
+    path.join(packageRoot, 'src/scripts/ai_council/learn_low_impact_preview.ts'),
+);
+
+// Both paths are RELATIVE and must be resolved explicitly — the corpus
+// against the project root, the seed against the PACKAGE root. Passed bare
+// they resolve against cwd, where a consumer project carrying a file at the
+// seed's own path would be read as the package seed and silently mark its
+// entries `already_seeded`.
+const plan = build_preview(
+    path.join(projectRoot, 'agents/decisions/low-impact-decisions.md'),
+    path.join(packageRoot, PACKAGE_SEED_PATH),
+    {
+        repoSlug: '<owner>/<repo>', // from `git remote get-url origin`
+        repoRoot: '<absolute repo root>',
+        privateDomains: [], // from .agent-settings.yml policy
+        customerNames: [], // from .agent-settings.yml policy
+        sqlIdentifiers: [], // from .agent-settings.yml policy
+    },
+);
 ```
 
+`PACKAGE_SEED_PATH` is the one place a CALLER reads that literal from, so no
+caller restates it; the privacy floor names the same path to lock it, which is
+a lock rather than a second source. The file it names is not in the tree, and
+that is not an error: the builder returns an empty already-seeded set, which
+is the current state of this package.
+
 The builder runs all three contract checks in one pass:
 
-1. Parses `## Validated` (strict mode — drift surfaces as
-   `CorpusParseError`).
+1. Parses `## Validated` (strict mode — drift surfaces as a parse error).
 2. Diffs against the seed file — already-seeded entries land in
-   `plan.already_seeded` and never upstream.
-3. Re-runs `redact_low_impact_entry` on every candidate. Failures
-   land in `plan.refused`.
+   `plan.already_seeded`.
+3. Re-runs `redact_low_impact_entry` on every candidate. Failures land in
+   `plan.refused`.
 
 ### 2. Surface the preview block
 
-Print `plan.render()` verbatim. Always. This is the user-facing
-audit trail per `fast-path-marker-visibility` Iron Law — the host
-agent MUST NOT swallow or paraphrase it.
+Print `plan.render()` verbatim. Always. This is the user-facing audit trail
+per the `fast-path-marker-visibility` Iron Law — the host agent MUST NOT
+swallow or paraphrase it.
 
 ```
 ## learn-low-impact preview — repo=<slug>
@@ -92,57 +111,25 @@ seed: <path>
 ### Already seeded (K) …
 ```
 
-### 3. Decide based on the flag
-
-- **`--preview` (default)** — stop here. If `plan.would_open_pr` is
-  true, the rendered block ends with
-  `> Re-run with \`--apply\` to open the draft PR via \`upstream-contribute\`.`
-  Hand control back to the user.
-- **`--apply`** — refuse when `plan.refused` is non-empty; surface
-  the refusals and stop. Otherwise invoke
-  [`upstream-contribute`](../skills/upstream-contribute/SKILL.md)
-  with:
-    - **target file:** `.agent-src.uncondensed/data/low-impact-decisions-seed.md`
-    - **PR title:** `plan.render_pr_body()` first heading
-    - **PR body:** `plan.render_pr_body()`
-    - **patch:** `plan.render_diff()`
-    - **draft:** `true` — never auto-merge; review is a human gate.
-
-### 4. Advance the local baseline (`--apply` path only)
-
-When the PR is opened (step 3 returns a PR URL + new commit SHA on
-the package branch):
-
-```bash
-# update the local pointer so subsequent runs are deltas
-sed -i.bak -E "s|^last-upstreamed: .*|last-upstreamed: <new-sha>|" \
-    agents/decisions/low-impact-decisions.md
-rm agents/decisions/low-impact-decisions.md.bak
-```
+### 3. Stop
 
-### 5. Surface result (`--apply` path only)
-
-```
-> Drafted PR <url>
-> Entries upstreamed: N
-> Provenance bumped: <old-sha> → <new-sha>
-```
+Hand control back to the user. The block is the deliverable. Nothing in this
+command sends anything; if the package ever ships a seed corpus again, the
+`revisit if` on decision D2 of
+`road-to-one-upstream-that-checks-what-leaves` fires and the route is designed
+once, on the one ladder.
 
 ## Halt conditions
 
-- Redactor refuses (any class) — surface, stop. `--apply` is rejected.
-- No candidates (`plan.has_work == False`) — exit 0 with the preview
-  block; no PR even on `--apply`.
-- `--preview` (default) — always stops before any side-effect.
-- Package repo unavailable per `upstream-contribute § Step 4` — surface
-  the access options, stop.
-- User explicitly declines the PR option — exit 0, no PR.
+- Redactor refuses (any class) — surface the refusals in the block, stop.
+- No candidates (`plan.has_work` is false) — exit 0 with the preview block.
+- Corpus missing or drifted — surface the parse error, stop.
 
 ## See also
 
-- [`upstream-contribute`](../skills/upstream-contribute/SKILL.md) — PR
-  machinery (branch, commit, gates).
 - [`agents/decisions/low-impact-decisions.md`](../../agents/decisions/low-impact-decisions.md)
   — the project-local corpus.
 - [`low-impact-corpus-privacy-floor`](../rules/low-impact-corpus-privacy-floor.md)
   — Iron Law.
+- [`upstream-contribute`](../skills/upstream-contribute/SKILL.md) — the one
+  front on the one outward step, for findings that are worth sending.

package/dist/agent-src/commands/memory/propose.md

@@ -43,6 +43,18 @@ Do NOT use when:
 
 ## Steps
 
+### 0. Show what a retro already captured
+
+```bash
+./scripts-run src/scripts/retro learnings
+```
+
+Lists every retro learning still waiting for a person — proposal files under
+`agents/proposals/retro-*.md` and intake signals with `source: retro` — each with
+the unit key of the finished work it came from. Show the list when it is not
+empty: a finding the user is about to propose may already be one of them, and
+promoting a retro learning is this person's act, never the retro's.
+
 ### 1. Pick the memory type
 
 ```

package/dist/agent-src/commands/memory.md

@@ -35,7 +35,7 @@ commands with a single entry point + sub-command dispatch.
 | `/memory mine-session` | `commands/memory/mine-session.md` | Mine the active session transcript for memory signals (preview-by-default) |
 | `/memory promote` | `commands/memory/promote.md` | Promote an intake signal to a curated memory entry |
 | `/memory propose` | `commands/memory/propose.md` | Append a provisional signal to the intake stream |
-| `/memory learn-low-impact` | `commands/memory/learn-low-impact.md` | Draft an upstream PR from the project's low-impact decision corpus (redactor-gated) |
+| `/memory learn-low-impact` | `commands/memory/learn-low-impact.md` | Preview which low-impact decisions would be worth upstreaming (redactor-gated, read-only) |
 
 Sub-command names match the locked contract in
 [`docs/contracts/command-clusters.md`](../docs/contracts/command-clusters.md).

package/dist/agent-src/commands/optimize.md

@@ -5,7 +5,7 @@ disable-model-invocation: true
 argument-hint: "[skills|agents-dir|augmentignore|rtk|project|prompt|deep] [args]"
 pack: meta
 intent: "Optimization dispatcher — skills, rtk, augmentignore, agents-dir, project sweep, prompt, deep autonomous loop"
-routes_to: [optimize-skills, optimize-rtk, optimize-augmentignore, optimize-agents-dir, optimize-project, optimize-prompt, optimize-deep]
+routes_to: [optimize-skills, optimize-rtk, optimize-augmentignore, optimize-agents-dir, optimize-project, optimize-prompt, analyze-and-optimize]
 replaces: []
 visibility: advanced
 description: Optimize orchestrator — routes to skills, agents-dir, augmentignore, rtk-filters, project (project-wide sweep), prompt (AI-prompt polish), deep (autonomous deep-refactoring loop)
@@ -41,7 +41,7 @@ the **agent layer** (skills, agents-dir, augmentignore, rtk), the **project**
 | `/optimize rtk` | `commands/optimize/rtk.md` | Create or refine project-local rtk filters |
 | `/optimize project` | `commands/optimize/project.md` | Project-wide optimization sweep — inventory roadmaps/ADRs/agent folders, challenge stale decisions, emit new roadmap(s) |
 | `/optimize prompt` | `commands/optimize/prompt.md` | Optimize a raw prompt for ChatGPT/Claude/Gemini via the 4-D methodology |
-| `/optimize deep` | `commands/optimize/deep.md` | Autonomous deep-refactoring loop — subagent analysis, council, central + sub-roadmaps, PR, N refinement loops (default 3) |
+| `/optimize deep` | `commands/analyze/and-optimize.md` | Alias of `/analyze:and-optimize` (ADR-288) — the autonomous deep-improvement loop: analysts, council, central + sub-roadmaps, PR, N refinement loops (default 3) |
 
 Sub-command names match the locked contract in
 [`docs/contracts/command-clusters.md`](../docs/contracts/command-clusters.md).
@@ -50,7 +50,8 @@ Sub-command names match the locked contract in
 
 1. Parse the user's argument: `/optimize <sub-command> [args]`.
 2. Look up the sub-command in the table above.
-3. Load the body of the corresponding `commands/optimize/<sub>.md` file and
+3. Load the body of the file the table names (`commands/optimize/<sub>.md`;
+   `deep` is the alias of `commands/analyze/and-optimize.md`) and
    follow its `## Steps` (or `## Instructions`) section verbatim.
 4. If the sub-command is unknown or missing, print the menu and ask:
 

package/dist/agent-src/commands/post-as/me.md

@@ -96,6 +96,24 @@ Generate the body as a single fenced markdown block per
 the loaded fingerprint (sentence-length ±25 %, pace),
 ±15 % length tolerance, write in `language`.
 
+### 5b. Check the finished draft
+
+Pipe the step-5 block on stdin — this command never writes the draft to
+disk; the checker scans inside a single wrapping fence:
+
+```bash
+npx tsx node_modules/@event4u/agent-config/src/scripts/check_voice_lexicon.ts --draft - \
+  --user .agent-user.md [--slot agents/context-spine/brand-voice.md]
+```
+
+Pass `--slot` only when `agent-config brand:status --json` reports the
+`brand-voice` slot valid. Claim phrases are checked either way; `--user`
+reads the user's own optional `style.avoid` (absent → no user terms).
+Findings flag without blocking: print each after the fenced
+block with its line and its note or the missing proof. Exit 2 is a broken
+invocation: say so and never present the draft as checked. No runtime → say
+so in one line.
+
 ### 6. Disclosure footer (omitted — user is the author)
 
 **Do not append a footer.** `/post-as:me` is the self-author path;