@event4u/agent-config 16.2.0 → 16.4.0

package/dist/agent-src/skills/tailwind-engineer/SKILL.md

@@ -87,7 +87,7 @@ Structure, controls, grid and breakpoints are never this skill's to adjust;
 those stay 1:1 with the artifact and belong to the rule, not to a
 utility-class decision. **Icons are out of scope here as well, and this skill
 asserts no obligation about them** — the icon system belongs to
-[`icon-consistency`](../../rules/icon-consistency.md). An earlier revision of
+`icon-consistency`. An earlier revision of
 this paragraph claimed icons "stay 1:1 with the artifact" with no citation of
 any kind, while putting them outside its own scope in the same sentence; the
 claim is withdrawn rather than defended, because a clause that reached the tree
@@ -97,7 +97,9 @@ Token authoring (DTCG 3-layer model, CSS-var/Tailwind generation) lives
 in [`design-tokens`](../design-tokens/SKILL.md); its
 `tokens.ts validate --dir <path>` is the **single token-discipline
 linter** — the mechanical check behind this rule (no hardcoded hex / px /
-rem outside the token files). Greenfield Tailwind config: bundled
+rem outside the token files). That skill ships with the `frontend-design`
+pack: without it the rule above still binds and is checked by reading,
+not by a linter. Greenfield Tailwind config: bundled
 `scripts/tailwind_config_gen.ts` (Apache-2.0-derived, pure templating)
 scaffolds `tailwind.config.{js,ts}` per framework.
 

package/dist/agent-src/skills/test-case-discovery/SKILL.md

@@ -39,6 +39,13 @@ CAP PER BEHAVIOR: 5–8 CASES — EACH MUST FAIL FOR A DISTINCT REASON.
 
 ## Does this change owe an executable behavior contract?
 
+This section is the decision *Gherkin*, *BDD*, *Cucumber*, *Behat* and
+*given-when-then* are meant to reach — each pinned as a should-trigger case in
+`evals/triggers.json`, with German exemplars for *Cucumber* and *Behat*. That
+corpus specifies
+the intent; it does not route, and this skill's `description` carries none of
+the terms. A decision, not a guide to writing one.
+
 Cases and contracts are not the same artifact. A **case** is something you assert
 in the project's unit runner. An **executable behavior contract** — a
 specification written in the language of whoever asked for the change, and run

package/dist/agent-src/skills/test-driven-development/SKILL.md

@@ -313,6 +313,23 @@ export async function retry<T>(op: () => Promise<T>): Promise<T> {
 Run again → passes. Configurable attempt count, backoff, and jitter all
 wait for their own failing tests.
 
+## The filtered probe, per ecosystem
+
+The discipline above is stack-independent; the **command** is not, and every
+runner invocation elsewhere in this file is a PHP one. Resolve the runner with
+`resolve_toolchain` (`work_engine/stack/runner.ts`) and read its `ecosystems`,
+then filter with that ecosystem's own form — never the whole suite:
+
+| `ecosystems` | Filtered probe |
+|---|---|
+| `php` | `vendor/bin/pest --filter '<name>'` · `php artisan test --filter '<name>'` |
+| `js` | `npx vitest run -t '<name>'` · `npx jest -t '<name>'` |
+| `python` | `pytest -k '<expr>'` · `pytest path/to/test.py::test_name` |
+| `go` | `go test -run '<regexp>' ./pkg/...` |
+
+A TypeScript repository resolves the ecosystem `js`, never `typescript`. The
+whole suite is the final gate and never the per-iteration probe, in every row.
+
 ## Gotchas
 
 * Running the full suite instead of a filtered test hides the RED→GREEN

package/dist/agent-src/skills/test-performance/SKILL.md

@@ -174,4 +174,22 @@ Replace dynamic `getPdo()` probing with explicit config:
 - Do NOT use RefreshDatabase when DatabaseTransactions works — 10x slower.
 - Do NOT run full test suite on every code change — use `--filter`.
 - Do NOT add test-only indexes — fix test design instead.
-- Do NOT disable parallel testing to "fix" flaky tests — fix the root cause.
+- Do NOT disable parallel testing to "fix" flaky tests — fix the root cause.
+
+## Where the time goes, per ecosystem
+
+The diagnosis above is PHPUnit-shaped. The method — measure before optimising,
+find the slowest tests first, separate setup cost from assertion cost —
+transfers; the instrument does not. Resolve the runner with
+`resolve_toolchain` and read its `ecosystems`:
+
+| `ecosystems` | Slowest-test instrument | The usual first cause |
+|---|---|---|
+| `php` | `--profile` / Pest's duration column | database refresh per test |
+| `js` | `vitest --reporter=verbose`, `jest --detectOpenHandles` | module transform and a leaked handle |
+| `python` | `pytest --durations=20` | session-scoped fixtures rebuilt per test |
+| `go` | `go test -v`, `-cpuprofile`, `-race` for the slow-and-flaky pair | a `TestMain` that starts real I/O |
+
+`go test` caches a package's result until its inputs change, so a suite that
+looks fast on the second run may not have run at all — `-count=1` defeats the
+cache when you are measuring.

package/dist/agent-src/skills/testing-anti-patterns/SKILL.md

@@ -206,7 +206,7 @@ Backstop grep — skip/xfail added to chase green (a hit in a diff that also "fi
 rg -n '\b(it|test|describe)\.skip\b|\.only\b|xit\(|xdescribe\(|@pytest\.mark\.(skip|xfail)|@Disabled|->markTestSkipped\(' .
 ```
 
-This is the test-surface instance of the `autonomous-execution` N=3 / allowlist-growth antipattern (bulk `skip`/`xfail` to force green counts as the tool being wrong, not the content) and the `verify-before-complete` floor (green must be *earned*, not manufactured).
+This is the test-surface instance of the `autonomous-execution` allowlist-growth antipattern (bulk `skip`/`xfail` to force green counts as the tool being wrong, not the content) and the `verify-before-complete` floor (green must be *earned*, not manufactured).
 
 ### The named smells — canonical vocabulary, and what it does NOT cover
 

package/dist/agent-src/skills/ui-component-architect/SKILL.md

@@ -61,7 +61,7 @@ than under-splitting. Meet BOTH conditions before you extract.
 **Read `DESIGN.md` § Owned components first** — it is the project's own inventory of what
 it already owns (component, status, story file, registry item), filled from the story files
 rather than from memory by
-[`design-system-capture`](../design-system-capture/SKILL.md). It is the cheapest possible
+`design-system-capture`. It is the cheapest possible
 prior-art check: one table, and a `deprecated` row tells you not to reach for something that
 still exists in the tree. Absent or empty → fall through to
 [`existing-ui-audit`](../existing-ui-audit/SKILL.md), which reads the codebase itself.
@@ -104,7 +104,7 @@ checkable before it reaches a page. That skill is stack-agnostic and ships in
 `engineering-base`, so a plain-`react` project receives it exactly as a
 `react-shadcn` one does. The state-coverage matrix the story set is derived
 from is React-specific and lives in
-[`react-shadcn-ui`](../react-shadcn-ui/SKILL.md) § Step 3, deliberately not
+`react-shadcn-ui` § Step 3, deliberately not
 duplicated here; a Blade, Vue or Livewire lane derives its story set from that
 lane's own state list instead.
 

package/dist/agent-src/skills/using-git-worktrees/SKILL.md

@@ -136,7 +136,7 @@ passes. For the global location, skip — path is outside the repo.
 ### 4. Create the worktree
 
 ```bash
-git worktree add .worktrees/<branch-name> -b <branch-name>
+git worktree add ".worktrees/<branch-name>" -b "<branch-name>"
 ```
 
 **Then address the worktree by flag, not by `cd`.** Every tool that matters

package/dist/agent-src/skills/verify-repair-loop/SKILL.md

@@ -75,7 +75,7 @@ regressions are detectable. Pick the verdict commands per the project's
 |---|---|---|
 | `threshold` | all targeted checks green | numeric pass bar, **absolute** (fraction of total), not "± N tests" |
 | `allow_regressions` | `false` | a revision breaking a baseline-green check stops the loop |
-| `max_attempts` | **3** | hard cap (test feedback is more actionable than diff critique → one more than `do-and-judge`'s 2; bounded by [`autonomous-execution`](../../rules/autonomous-execution.md) N=3) |
+| `max_attempts` | **3** | hard cap, this loop's own number (test feedback is more actionable than diff critique → one more than `do-and-judge`'s 2); independent of [`autonomous-execution`](../../rules/autonomous-execution.md)'s `execution.fix_loop_max` budget, which now defaults to 10 |
 | `plateau_window` | 3 | stop if the last *window* scores sit within `tolerance` |
 | `tolerance` | small absolute fraction | flake jitter that does **not** count as improvement |
 
@@ -294,7 +294,7 @@ iter2 fixes A, breaks B, 92% → "PASS"      ✗ regression ignored
   phantom-verification check bound at step 3.4, with its fixture corpus in
   `tests/fixtures/tamper-corpus/` and its polarity proven both ways by
   `--self-test`.
-* [`autonomous-execution`](../../rules/autonomous-execution.md) — the N=3 cap.
+* [`autonomous-execution`](../../rules/autonomous-execution.md) — the `execution.fix_loop_max` cap (this loop's own `max_attempts` is a separate, smaller number).
 * [`resident-process-governance`](../../docs/contracts/resident-process-governance.md) — the
   runtime-free constraint this loop honors.
 * Optional impact pre-step (when a code-graph is present, ADR-124): run

package/dist/agent-src/templates/AGENTS.md

@@ -14,7 +14,7 @@
 ## Pointers
 
 - **Filling out this AGENTS.md** — section templates, capability bullets, multi-agent entry flow, monorepo layout: [`consumer-agents-md-guide`](.augment/contexts/contracts/consumer-agents-md-guide.md).
-- **Behavior rules** (9 always-on; rest routed) — [`.augment/rules/`](.augment/rules/). Kernel rules are **immutable**; tighten only via the override registry.
+- **Behavior rules** (9 always-on; rest routed) — [`.augment/rules/`](.augment/rules/). Kernel rules change only by a ratified edit; overrides may tighten.
 - **Skills (on-demand expertise)** — domain skills surfaced by description, invoked on trigger match: [`.augment/skills/`](.augment/skills/).
 - **Commands (workflows)** — slash-commands the agent runs end-to-end (`/work`, `/create-pr`, …): [`.augment/commands/`](.augment/commands/).
 - **Project-specific docs** — your own architecture notes, roadmaps, sessions, contexts: [`agents/`](agents/).

package/dist/agent-src/templates/agent-settings.md

@@ -111,20 +111,6 @@ personal:
   # Detected and set by /onboard on first run.
   rtk_installed: false
 
-  # Minimal output mode (true, false)
-  # true  = short bullet points during work, concise summary at the end
-  # false = verbose explanations and reasoning
-  minimal_output: true
-
-  # Play-by-play mode (true, false)
-  # true  = briefly share intermediate findings as you go
-  # false = silently investigate, only report the conclusion
-  play_by_play: false
-
-  # Prefix PR comment replies with a bot icon 🤖 (true, false)
-  # Personal preference — each developer decides for themselves.
-  pr_comment_bot_icon: false
-
   # Autonomous execution — suppress trivial workflow questions (on, off, auto)
   # on   = act on the obvious next step; never ask "Step 2 or 3?", "should I commit?", etc.
   # off  = ask trivial workflow questions (legacy behavior)
@@ -136,9 +122,6 @@ personal:
 
 # --- Project / team preferences ---
 project:
-  # Path to the PR template file (relative to project root)
-  pr_template: .github/pull_request_template.md
-
   # Target repository for universal improvement PRs (e.g. org/agent-config)
   upstream_repo: ""
 
@@ -224,14 +207,6 @@ hooks:
     # outside the standard location.
     # script: scripts/chat_history.ts
 
-# --- Optional pipelines ---
-pipelines:
-  # Skill improvement pipeline (true, false)
-  # true  = after meaningful tasks, propose learning capture and improvements (default)
-  # false = silent, no post-task analysis
-  # Included by every rule_loading_tier except `custom`.
-  skill_improvement: true
-
 # --- Roadmap execution ---
 #
 # Controls when /roadmap:process-* runs the project's quality pipeline.
@@ -239,22 +214,6 @@ pipelines:
 # response — that cadence is governed by `roadmap-progress-sync` and
 # is non-negotiable. This setting only governs *quality tool runs*.
 roadmap:
-  # Skip the pre-run confirmation gate when /roadmap:process-step|phase|full
-  # is invoked (true, false).
-  #   true  = default. The command name itself names the scope; if a
-  #           single active roadmap is discoverable (or the user named
-  #           one), the loop starts immediately. The resolved roadmap /
-  #           cadence / council setting are surfaced inline as a one-shot
-  #           note so an unwanted pick can still be aborted mid-stream.
-  #   false = legacy. The loop shows the pre-run summary with numbered
-  #           options (Go / Different roadmap / Different scope / Toggle
-  #           council / Abort) and waits for input.
-  # The gate is still shown — regardless of this flag — when the roadmap
-  # is ambiguous (multiple active, none named) or a scope / cadence
-  # conflict has no sensible default. The flag suppresses the
-  # confirmation, never a genuine "which roadmap?" question.
-  skip_pre_run_gate: true
-
   # When to run quality tools during /roadmap:process-step|phase|full.
   #   end_of_roadmap = once, before archiving (default — fastest, fewest tokens)
   #   per_phase      = once after every completed phase
@@ -292,7 +251,8 @@ subagents:
   judge_model: ""
 
   # Maximum number of parallel subagent invocations (integer, default 3)
-  # Set to 1 to serialize. Hard cap enforced by runtime.
+  # Set to 1 to serialize. A limit the model reads from settings — no
+  # code counts spawns against it.
   max_parallel: 3
 
 # --- Recursive self-verification (see skills/recursive-verification) ---
@@ -369,17 +329,8 @@ onboarding:
 # never auto-executes — the user picks. See `rules/command-suggestion-policy.md`.
 commands:
   suggestion:
-    # Master switch (true, false). `false` = the layer is silent;
-    # explicit `/commands` still work as today.
-    enabled: true
-    # Minimum match score (0.0–1.0) before a suggestion surfaces.
-    confidence_floor: 0.6
-    # Cooldown in seconds between re-suggestions of the same
-    # (command, evidence) pair. Default 600 = 10m.
-    cooldown_seconds: 600
-    # Max number of command suggestions before the as-is option.
-    # The as-is option is always extra (total rendered = max_options + 1).
-    max_options: 4
+    # The layer is always on; floor 0.6, cooldown 600s and a cap of 4
+    # options (plus the as-is option) are fixed.
     # Commands to never suggest. Still work when typed explicitly.
     blocklist: []
 
@@ -398,23 +349,13 @@ commands:
   # Critical info (breaking changes, migrations, security, rollback) is
   # ALWAYS included at every tier — the tier governs explanatory depth only.
   #
-  # api_examples (default true): add a fenced JSON example for API-endpoint
-  #   changes ONLY when grounded in a real source (DTO/resource, OpenAPI,
-  #   fixture, probe); else a one-line pointer, never an invented example.
-  #
   # screenshots (default false): capability-gated frontend screenshots.
   #   true attempts capture only when the host has browser/preview tooling;
   #   emits a note and leaves the placeholder when absent — never blocks the PR.
-  #
-  # ui_paths / api_paths: optional globs to make frontend / API detection
-  #   explicit; empty = a light heuristic that fails open.
   create_pr:
     preview_description: false
     detail_level: min
-    api_examples: true
     screenshots: false
-    ui_paths: []
-    api_paths: []
 
 # --- Telemetry (artefact engagement, default-off) ---
 #
@@ -426,36 +367,20 @@ commands:
 # (once Phase 3 of road-to-artifact-engagement-telemetry lands).
 # --- Verbosity (token frugality) ---
 #
-# Five toggles controlling what the agent shows after acting.
-# Default = terse. Flip to true to restore legacy verbose output.
+# What the agent shows after acting. Default = terse. Four former toggles
+# (preview_artifacts, routine_confirmations, post_action_reports,
+# intent_announcements) were retired on 2026-10-09 with the terse default
+# as fixed behavior.
 # See agents/roadmaps/road-to-token-frugality.md for the full rationale
 # and the contexts/contracts/frugality-charter.md for the writer-side
 # standard.
 verbosity:
-  # Show generated commit messages, PR titles/bodies, branch names
-  # before acting. false = use generated content directly.
-  preview_artifacts: false
-
-  # Confirmation prompts for routine workflow steps when there is
-  # one obvious answer ("looks good — commit?"). Iron-Law gates
-  # (commit-policy, scope-control git-ops, non-destructive) ALWAYS
-  # ask regardless of this flag.
-  routine_confirmations: false
-
   # Offer "run AI Council on this?" inside delivery commands
   # (/feature-plan, /review-changes, /roadmap-create). Council
   # commands themselves (/council, /create-pr → already excluded)
   # are unaffected.
   offer_council_in_delivery: false
 
-  # Multi-line status / summary blocks after a successful action.
-  # off | minimal | full — default minimal (one-line confirmation).
-  post_action_reports: minimal
-
-  # Intent announcements ("Let me check…", "Now I will…", "Found
-  # it") in skill bodies. false = act and emit the result.
-  intent_announcements: false
-
   # Script stdout chatter from `scripts/*.ts`, `scripts/*.sh`, and
   # `.augment/scripts/`. Read by the helper module
   # `scripts/_lib/script_output.ts`.
@@ -473,19 +398,6 @@ verbosity:
   # safe task per Phase 10.3.
   taskfile_command_echo: false
 
-# --- Telegraph condensation (output-side) ---
-#
-# One lever, deliberately: whether the telegraph-speak rule ships at all.
-# The rule body states its own grammar scope — which carve-outs (numbered
-# options, Iron-Law literals, code, paths) stay full prose — so there is no
-# separate scope key to pin.
-telegraph:
-  # speak: false = DORMANT (the shipped default). The rule is omitted from
-  # dist/router.json entirely, so its body never reaches a host. That
-  # omission is what stops the cost. Set true only after an output-side
-  # bench clears the kill-criterion bar.
-  speak: false
-
 telemetry:
   artifact_engagement:
     # Master switch. `false` (default) produces zero file IO and zero
@@ -612,11 +524,7 @@ the canonical narrative lives in
 | `personal.open_edited_files` | `true`, `false` | `false` | Auto-open edited files in the IDE after edits |
 | `personal.canary_name` | first name | _(empty)_ | Per-project OVERRIDE of the name the agent addresses the user with (session-canary). The name itself lives user-globally: `identity.name` in the wizard's `settings/.agent-user.yml`, fallback `personal.canary_name` in the user-global settings. See `rules/session-canary.md`. |
 | `personal.rtk_installed` | `true`, `false` | `false` | Whether rtk (Rust Token Killer) is installed. Detected and set by `/onboard`. |
-| `personal.minimal_output` | `true`, `false` | `true` | When `true`: short bullet points during work, concise summary at end. When `false`: verbose explanations. |
-| `personal.play_by_play` | `true`, `false` | `false` | When `true`: share intermediate findings during investigation. When `false`: work silently, report only the conclusion. |
-| `personal.pr_comment_bot_icon` | `true`, `false` | `false` | Prefix PR comment replies with 🤖 to indicate bot-authored replies. Personal preference — each developer decides. |
 | `personal.autonomy` | `on`, `off`, `auto` | `auto` | Suppress trivial workflow questions and act on the obvious next step. `auto` (default) defaults to `off` but flips to `on` after a prose opt-in like "arbeite selbstständig". `on` suppresses trivial questions unconditionally. `off` is the legacy ask-everything mode. Blocking decisions (security, scope expansion, push/merge/branch/PR/tag) are never suppressed. See `rules/autonomous-execution.md`. |
-| `project.pr_template` | file path | `.github/pull_request_template.md` | Path to PR template file. Read this instead of searching for it. |
 | `project.upstream_repo` | `org/repo` | _(empty)_ | Target repository for universal improvement PRs (e.g., `org/agent-config`). |
 | `project.improvement_pr_branch_prefix` | string | `improve/agent-` | Branch prefix for agent improvement PRs. |
 | `github.pr_reply_method` | `replies_endpoint`, `create_review_comment`, `auto` | `create_review_comment` | GitHub API method for replying to PR review comments. `auto` detects on first use. |
@@ -625,12 +533,10 @@ the canonical narrative lives in
 | `chat_history.enabled` | `true`, `false` | `true` | Persist chat events to `agents/runtime/.agent-chat-history` (JSONL) for crash recovery. |
 | `chat_history.frequency` | `per_turn`, `per_phase`, `per_tool` | per profile | Logging granularity. Defaults: `minimal`→`per_turn`, `balanced`→`per_phase`, `full`→`per_tool`. |
 | `chat_history.text_limits.{user,agent,tool,phase}` | integer (chars) | `user=0`, `agent=5000`, `tool=200`, `phase=200` | Per-entry-type text-length cap. `0` = verbatim, no slice. `N > 0` = collapse whitespace, slice to N chars, append `" … [+K chars]"` so the log self-reports truncation. Defaults match `DEFAULT_TEXT_LIMITS` in `scripts/chat_history.ts`. |
-| `hooks.concern_budget.{max_per_event,tier1_concerns,hard_fail}` | integer / list / `true`,`false` | `8` / `[]` / `false` | Concern budget gate for the hook dispatcher (`lint_hook_concern_budget`): caps concerns per (platform, event) cell, restricts `fail_closed: true` to the `tier1_concerns` allowlist, and `hard_fail: false` keeps the gate warn-only. |
+| `hooks.concern_budget.tier1_concerns` | list | `[]` | Concern budget gate for the hook dispatcher (`lint_hook_concern_budget`): restricts `fail_closed: true` to this allowlist. The per-cell cap (8) is fixed and the gate is warn-only unless run with `--strict`. |
 | `hooks.injection_scan.enabled` | `true`, `false` | `false` | PostToolUse prompt-injection scanner: scans tool output for injection signatures and warns in context (exit 2) — never blocks. |
 | `hooks.rtk_wrap.enabled` | `true`, `false` | `false` | PreToolUse RTK-wrap nudge: when `rtk` is on PATH, warns (exit 2, never blocks) to re-run a verbose CLI command wrapped with rtk. |
 | `hooks.design_slop.enabled` | `true`, `false` | `false` | PreToolUse anti-slop nudge: runs the `lint_design_slop` registry against about-to-be-written UI content and warns (exit 2, never blocks) on P0/P1 aesthetic tells. |
-| `pipelines.skill_improvement` | `true`, `false` | `true` | When `true`: propose learning capture after meaningful tasks. When `false`: silent. Included in every profile except `custom`. |
-| `roadmap.skip_pre_run_gate` | `true`, `false` | `true` | When `true` (default): `/roadmap:process-step\|phase\|full` skips the interactive pre-run summary and starts the loop immediately — the resolved roadmap, cadence, and council are surfaced inline so an unwanted pick can still be aborted. When `false`: the loop shows the pre-run summary with numbered options (Go / Different roadmap / Different scope / Toggle council / Abort) and waits. The gate is always shown — regardless of this flag — when the roadmap is ambiguous (multiple active, none named) or a scope / cadence conflict has no sensible default. |
 | `roadmap.quality_cadence` | `end_of_roadmap`, `per_phase`, `per_step` | `end_of_roadmap` | When `/roadmap:process-step|phase|full` runs the project's quality pipeline — only relevant when `quality.local_auto_run` is `true`; when it is `false` (the default) local pipeline runs are suppressed at every cadence and remote CI is the gate. Default skips per-step / per-phase runs and gates only the final archival. `per_phase` runs once after every phase; `per_step` is the legacy verbose mode. Step checkboxes and the dashboard are always updated regardless. |
 | `quality.local_auto_run` | `true`, `false` | `false` | When `false` (default): agent NEVER runs the project's quality pipeline (`task ci`, `make test`, `npm run check`, PHPStan, ECS, Rector, test suites) proactively — and does not ask. The user runs quality tools manually; remote CI is the authoritative gate. The agent runs one only on (1) an explicit ask this turn, (2) a concrete CI failure (run exactly that failing check), or (3) the new-gate carve-out. When `true`: opt-in legacy — agent runs the pipeline autonomously when work is ready for verification. **Carve-out**: NEW CI gates / smoke tests / test files MUST run locally regardless of this flag — without execution the new gate is unverified evidence. Iron Law `verify-before-complete` still applies; suppressed runs are surfaced ("quality gates delegated to remote CI"), never claimed as passing. |
 | `subagents.implementer_model` | model alias or empty | _(empty)_ | Model for implementer subagents. Empty = same tier as session model. See [subagent-configuration](../contexts/subagent-configuration.md). |
@@ -641,23 +547,11 @@ the canonical narrative lives in
 | `personas.override` | list of persona ids | `[]` | Developer-local override of the team default lens cast (not in the shipped personal template — add the block when needed). Empty = inherit `personas.default` from `.agent-project-settings.yml`. See [`layered-settings`](../guidelines/agent-infra/layered-settings.md). |
 | `personas.ignore` | list of persona ids | `[]` | Persona ids dropped from the default cast locally. Ignored personas stay invokable via `--personas=<id>`. |
 | `onboarding.onboarded` | `true`, `false` | `false` | Whether `/onboard` has run on this project. The `onboarding-gate` rule prompts for `/onboard` when this is `false`. Missing entirely = legacy project, treated as onboarded. |
-| `commands.suggestion.enabled` | `true`, `false` | `true` | Master switch for the command-suggestion layer. `false` = the layer is silent; explicit `/commands` still work. See `rules/command-suggestion-policy.md`. |
-| `commands.suggestion.confidence_floor` | `0.0`–`1.0` | `0.6` | Minimum match score before a suggestion surfaces. Per-command frontmatter (`suggestion.confidence_floor`) overrides this global floor. |
-| `commands.suggestion.cooldown_seconds` | integer | `600` | Cooldown between re-suggestions of the same `(command, evidence)` pair. `600` = 10m. |
-| `commands.suggestion.max_options` | integer | `4` | Max number of command suggestions before the always-present "run as-is" option (total rendered = `max_options + 1`). |
 | `commands.suggestion.blocklist` | list of command names | `[]` | Commands that never appear as a suggestion. They still work when typed explicitly. |
 | `commands.create_pr.preview_description` | `true`, `false` | `false` | When `false`: `/create-pr` skips the title/body preview + adjust loop and uses the generated content directly. Saves agent tokens. When `true`: show title and body before creating and ask for adjustments. `/create-pr:description-only` always previews regardless of this setting. |
 | `commands.create_pr.detail_level` | `min`, `med`, `max` | `min` | Verbosity tier of the PR Description body. `min` = title + 2-3 sentence what/why/impact + ticket; `med` = + grouped changes + tests note; `max` = + how-to-test + edge cases + reviewer guidance. Critical info (breaking / migration / security / rollback) is included at every tier. |
-| `commands.create_pr.api_examples` | `true`, `false` | `true` | Add a grounded JSON example for API-endpoint changes; `true` includes one only from a real source (DTO/OpenAPI/fixture/probe), else a pointer, never invented; `false` = never. |
 | `commands.create_pr.screenshots` | `true`, `false` | `false` | Capability-gated frontend screenshots. `true` attempts capture when browser/preview tooling is present; notes-and-skips otherwise, never blocks the PR. Before/after is best-effort. |
-| `commands.create_pr.ui_paths` | glob list | `[]` | Optional globs pinning frontend detection; empty = light heuristic, fail-open. |
-| `commands.create_pr.api_paths` | glob list | `[]` | Optional globs pinning API-endpoint detection; empty = light heuristic, fail-open. |
-| `verbosity.preview_artifacts` | `true`, `false` | `false` | Show generated commit messages, PR titles/bodies, branch names before acting. `false` = use generated content directly. See the token-frugality plate under `agents/roadmaps/` (Phase 2/3). |
-| `verbosity.routine_confirmations` | `true`, `false` | `false` | Confirmation prompts for routine workflow steps when there is one obvious answer ("looks good — commit?"). Iron-Law gates (`commit-policy`, `scope-control` git-ops, `non-destructive-by-default`) ALWAYS ask regardless. |
 | `verbosity.offer_council_in_delivery` | `true`, `false` | `false` | Offer "run AI Council on this?" inside delivery commands (`/feature-plan`, `/review-changes`). Council commands themselves are unaffected. `/roadmap:create` no longer reads it — under a mission the council is a step, resolved from `agent-config council:status`, never an offer. |
-| `verbosity.post_action_reports` | `off`, `minimal`, `full` | `minimal` | Multi-line status / summary blocks after a successful action. `off` = no report; `minimal` = one-line confirmation; `full` = bullet list. |
-| `verbosity.intent_announcements` | `true`, `false` | `false` | Intent announcements ("Let me check…", "Now I will…", "Found it") in skill bodies. `false` = act and emit the result. |
-| `telegraph.speak` | `true`, `false` | `false` | Whether the telegraph-speak rule ships at all. `false` (default) = dormant: the projector omits the rule from `dist/router.json`, so its body never reaches a host. The rule body states its own grammar scope and carve-outs (numbered options, Iron-Law literals, code, paths), so this is the only lever. |
 | `telemetry.artifact_engagement.enabled` | `true`, `false` | `false` | Master switch for the artefact engagement log. Not in the shipped template — a missing `telemetry:` section means disabled (the recording rule no-ops). Default-off; zero file IO and zero token cost when `false`. Maintainer-targeted; consumers leave it off. |
 | `telemetry.artifact_engagement.granularity` | `task`, `phase-step`, `tool-call` | `task` | Boundary at which events are recorded. `tool-call` is expensive — opt-in only. |
 | `telemetry.artifact_engagement.record.consulted` | `true`, `false` | `true` | When `true`: record artefacts loaded into context. |
@@ -675,11 +569,6 @@ the canonical narrative lives in
 | `linked_projects_max_files` | integer | `20000` | File-count ceiling above which a detected sibling is flagged `large` (awareness only). Never excludes. |
 | `knowledge.global_sharing.enabled` | `true`, `false` | `true` | **User-global** (keep in `~/.event4u/agent-config/agent-settings.yml`). Master switch for the file-first global knowledge-card store (ADR-100; default-ON per ADR-119, the council-validated bounded-downside flip superseding ADR-103 — adversarially spot-checked redaction incl. hidden-unicode hardening, narrowest tier default, pre-registered demotion trigger). `false` fully no-ops the layer (single-key revert); v1 project-local cards unaffected. |
 | `knowledge.global_sharing.allowed_tiers` | list of `public`,`vendor`,`proprietary` | `[public]` | Origin tiers auto-eligible to cross a project boundary. Narrowest default per ADR-119; `vendor` is a deliberate opt-in widening. `proprietary` is manual-only regardless (the gate hard-codes it). |
-| `knowledge.global_sharing.redaction.enabled` | `true`, `false` | `true` | Run the privacy-floor + source-confidentiality scan before any card goes global. |
-| `knowledge.global_sharing.redaction.halt_on_trigger` | `true`, `false` | `true` | Halt-and-surface on a confidential-pattern hit; never silent-share, never auto-rewrite. |
-| `knowledge.global_sharing.auto_promote_threshold` | integer | `2` | Distinct-repo count that triggers a promotion **suggestion** (never a silent write). |
-| `knowledge.global_sharing.freshness.hypothesis_after_days` | integer | `90` | Global card older than this is lead-only (positive structure re-confirmed before use). |
-| `knowledge.global_sharing.freshness.stale_after_days` | integer | `180` | Global card older than this is skipped until re-verified. |
 
 ### Rename-Map (migration)
 
@@ -693,15 +582,10 @@ Applied automatically when `scripts/install` finds a legacy `.agent-settings`
 | `open_edited_files` | `personal.open_edited_files` |
 | `user_name` | `personal.user_name` |
 | `rtk_installed` | `personal.rtk_installed` |
-| `minimal_output` | `personal.minimal_output` |
-| `play_by_play` | `personal.play_by_play` |
-| `pr_comment_bot_icon` | `personal.pr_comment_bot_icon` |
-| `pr_template` | `project.pr_template` |
 | `upstream_repo` | `project.upstream_repo` |
 | `improvement_pr_branch_prefix` | `project.improvement_pr_branch_prefix` |
 | `github_pr_reply_method` | `github.pr_reply_method` |
 | `eloquent_access_style` | `eloquent.access_style` |
-| `skill_improvement_pipeline` | `pipelines.skill_improvement` |
 | `subagent_implementer_model` | `subagents.implementer_model` |
 | `subagent_judge_model` | `subagents.judge_model` |
 | `subagent_max_parallel` | `subagents.max_parallel` |
@@ -709,6 +593,40 @@ Applied automatically when `scripts/install` finds a legacy `.agent-settings`
 Unknown keys in the legacy file are preserved under a `_legacy:` section
 so nothing is silently dropped; the migration log points them out.
 
+### Retired keys
+
+These keys were removed because their shipped default is the only behavior
+the package supports; the behavior they described is unchanged. A leftover
+value in an older `.agent-settings.yml` is ignored with one stderr line per key
+(`<key> was removed (<what decides instead>); ignored.`) — the registry is
+`REMOVED_KEYS` in `scripts/_lib/settings_removed_keys.ts`. Delete the line to
+silence it.
+
+| Retired 2026-10-09 | Fixed behavior |
+|---|---|
+| `personal.minimal_output`, `personal.play_by_play`, `verbosity.preview_artifacts`, `verbosity.routine_confirmations`, `verbosity.post_action_reports`, `verbosity.intent_announcements` | terse replies, no narration, no preview, no routine confirmation, one-line post-action report (`direct-answers` Iron Law 3) |
+| `personal.pr_comment_bot_icon` | no 🤖 prefix (`no-decorative-emojis-in-git-surfaces`) |
+| `telegraph.speak` | the telegraph-speak rule does not ship (bench verdict, ADR telegraph/0002) |
+| `tokens.rich_skills` | `token_budget_class: rich` skills load in full (`token-budget-discipline`) |
+| `reasoning.auto_gate`, `reasoning.components.*` (nine switches) | every RDP component follows the gate's task and host signals (`rdp-gate`); `reasoning.enabled` stays the hard off |
+| `roadmap.skip_pre_run_gate` | no pre-run confirmation; only an ambiguous roadmap or an unresolvable conflict shows the gate |
+| `roadmap.dashboard_regen_cadence` | dashboard regen every 5th closed step, at phase boundaries, at reply end and on any file-shape touch |
+| `commands.auto_detect` | orchestrators auto-detect; opt one out with `auto_detect: false` in its front-matter or one run with `--no-auto-detect` |
+| `commands.suggestion.enabled`, `.confidence_floor`, `.cooldown_seconds`, `.max_options` | suggestion layer always on, floor 0.6, cooldown 600 s, at most 4 options; `blocklist` and per-command frontmatter still apply |
+| `commands.create_pr.api_examples`, `.ui_paths`, `.api_paths` | grounded API examples always added; frontend / API detection by the light heuristic |
+| `memory.cadence` | the 🧠 memory line renders whenever a memory type was asked; `memory.visibility: off` still silences it |
+| `memory.review_threshold` | `/memory load` previews intake signals above a fixed 10 |
+| `knowledge.global_sharing.redaction.*`, `.auto_promote_threshold`, `.freshness.*` | redaction always runs and halts on a hit; promotion suggested at 2 repos; lead-only after 90 days, skipped after 180 |
+| `hooks.concern_budget.max_per_event`, `.hard_fail` | at most 8 concerns per (platform, event) cell; warn-only unless the gate runs with `--strict` |
+| `decision_engine.surface_traces` | no per-phase decision-trace file (accepted in the `decision_engine:` block and ignored) |
+| `decision_engine.on_block_fallback` | an `on_block: ask` that times out stops (accepted and ignored) |
+| `explain.enable_last` | `agent-config explain last` always renders the trace |
+| `project.pr_template` | `/create-pr` reads `.github/pull_request_template.md` |
+| `pipelines.skill_improvement` | post-task learning capture is always proposed (`skill-improvement-trigger`) |
+| `consistency.cross_source` | cross-source discrepancies are always surfaced and asked, batched into one question |
+| `subagents.downshift` | delegable sub-tasks always run on the lowest-capable tier they declare |
+| `ai_team.suppress_setup_hint` | the one-line team-mode setup hint is always shown (accepted and ignored in the `ai_team:` block) |
+
 ## Cost profiles
 
 The `rule_loading_tier` setting selects which agent surfaces are active. See
@@ -721,10 +639,9 @@ The `rule_loading_tier` setting selects which agent surfaces are active. See
 | `full` | `balanced` + Tool adapters (GitHub / Jira, read-only, opt-in). |
 | `custom` | Ignore profile — every matrix value must be set explicitly. |
 
-**Learning loop:** `pipelines.skill_improvement` is `true` by default and is
-included in every profile except `custom`. It triggers post-task learning
-capture via the `skill-improvement-trigger` rule. Flip to `false` in the
-settings file if you want a silent agent without touching the profile.
+**Learning loop:** post-task learning capture runs in every profile via the
+`skill-improvement-trigger` rule. It is a proposal the user confirms, not a
+setting (`pipelines.skill_improvement` was retired on 2026-10-09).
 
 Other per-feature toggles may be added in future releases; when they land,
 they ship with a live consumer in code and get documented here, not before.

package/dist/agent-src/templates/agents/agent-project-settings.example.yml

@@ -7,8 +7,8 @@
 # Precedence (lowest → highest):
 #   1. Package defaults (shipped by event4u/agent-config)
 #   2. ~/.event4u/agent-config/agent-settings.yml — user-global DX-comfort
-#      defaults (whitelist: name, ide, rule_loading_tier, personal.bot_icon,
-#      personal.autonomy, memory.cadence). Created on opt-in via
+#      defaults (whitelist: `MERGEABLE_KEYS` in scripts/_lib/agent_settings.ts,
+#      e.g. name, ide, rule_loading_tier, personal.autonomy). Created on opt-in via
 #      /onboard; project-local files always win. Legacy
 #      ~/.config/agent-config/agent-settings.yml is read as a fallback.
 #   3. This file (.agent-project-settings.yml) — team defaults
@@ -39,7 +39,7 @@ schema_version: 1
 # CI guard: a release bump of `package.json` must update this value
 # in lockstep — see scripts/check_template_pin_drift.ts (road-to-
 # portable-runtime-and-update-check P3.3).
-agent_config_version: "16.2.0"
+agent_config_version: "16.4.0"
 
 # --- Project identity ---
 project:

package/dist/agent-src/templates/command.md

@@ -74,8 +74,8 @@ suggestion:
      [Frugality Charter](../contexts/contracts/frugality-charter.md):
      no preview-then-confirm pair, no "Ready to proceed?" gate, no
      numbered options unless options differ in *consequence* (per
-     `no-cheap-questions § Pre-Send Self-Check`). Routine confirmations
-     are governed by `verbosity.routine_confirmations: false` (default).
+     `no-cheap-questions § Pre-Send Self-Check`). A routine step with one
+     obvious answer is never confirmed.
 
      Only emit a numbered-options block when ALL of:
        1. Two or more options carry distinct consequences (not sequencing/format),

package/dist/agent-src/templates/roadmaps.md

@@ -335,7 +335,12 @@ that was never a judgement call.
     same reply as the checkbox flip that cleared it, per
     [`roadmap-progress-sync`](../rules/roadmap-progress-sync.md). A
     step gated by a specific blocker may cross-reference it inline:
-    `- [ ] … <!-- blocked-by: <blocker-id> -->`.
+    `- [ ] … <!-- blocked-by: <blocker-id> -->`. A bare id resolves in the
+    same file only. A blocker declared in ANOTHER active roadmap or stub is
+    named `<!-- blocked-by: <roadmap-stem>#<blocker-id> -->`; a target found
+    only in `later/`, `archive/` or `skipped/` is not resolvable. The marker
+    must name an OPEN blocker — once it resolves, remove the marker — and
+    `lint_roadmap_blockers` reads it on a continuation line too.
 
     **A blocker stops execution; a hold stops publication.** The two are not
     interchangeable and never substitute for one another. A blocker says the work

package/dist/agent-src/templates/scripts/work_engine/_lib/agent_settings.ts

@@ -207,7 +207,6 @@ export const MERGEABLE_KEYS: readonly string[] = [
     // weak-host-lift-tiering + token-program-integration verdicts; ADR
     // records the whitelist addition). Wins over rule_loading_tier.
     'discipline_profile',
-    'memory.cadence',
     'personal.bot_icon',
     'personal.autonomy',
     // Knowledge-card global cross-project sharing is a USER-GLOBAL setting
@@ -215,11 +214,6 @@ export const MERGEABLE_KEYS: readonly string[] = [
     // ~/.event4u/agent-config/agent-settings.yml values are honoured.
     'knowledge.global_sharing.enabled',
     'knowledge.global_sharing.allowed_tiers',
-    'knowledge.global_sharing.redaction.enabled',
-    'knowledge.global_sharing.redaction.halt_on_trigger',
-    'knowledge.global_sharing.auto_promote_threshold',
-    'knowledge.global_sharing.freshness.hypothesis_after_days',
-    'knowledge.global_sharing.freshness.stale_after_days',
 ];
 
 const _DEFAULTS: SettingsDict = {};

package/dist/agent-src/templates/scripts/work_engine/directives/backend/_self_fix.ts

@@ -24,10 +24,13 @@
  *
  * Two floors, neither invented here:
  *
- * - **Ceiling** — {@link SELF_FIX_CEILING} attempts per lane, which is the
- *   N=3 validation-loop budget from the `autonomous-execution` rule, applied
- *   per validation target (that rule resets its counter on a different
- *   target, so the counters below are per-lane rather than per-run).
+ * - **Ceiling** — {@link SELF_FIX_CEILING} attempts per lane. This is the
+ *   engine's own number, independent of the `autonomous-execution` rule's
+ *   `execution.fix_loop_max` budget (now a default-10 ladder) — a
+ *   deterministic, critic-free retry on an already-known verdict has no
+ *   judgement call for that ladder's escalation rungs to apply to. The
+ *   counters below are per-lane rather than per-run; each lane is its own
+ *   target.
  * - **No-progress** — two consecutive attempts producing an identical verdict
  *   signature stop the loop immediately, budget remaining or not. This is the
  *   floor the `recursive-verification` skill already stated in prose ("two
@@ -47,9 +50,10 @@ import { type Any, Outcome, StepResult, agent_directive } from '../../delivery_s
 /**
  * Attempts allowed per lane before the loop exits PARTIAL.
  *
- * Deliberately the `autonomous-execution` N=3 budget rather than a fresh
- * number: a second, differently-sized retry ceiling in the same package would
- * be a rule the tree contradicts.
+ * The engine's own number — not borrowed from `autonomous-execution`'s
+ * `execution.fix_loop_max` budget, which now defaults to 10 with a
+ * strategy-shift ladder. A deterministic, critic-free retry on an
+ * already-known verdict has no judgement call that ladder's rungs apply to.
  */
 export const SELF_FIX_CEILING = 3;