@event4u/agent-config 9.9.0 → 9.10.0

package/dist/agent-src/commands/team-knowledge/bootstrap.md

@@ -22,9 +22,9 @@ packs:
 Seeds a fresh project's `agents/knowledge/` typed pages from
 **deterministic static analysis only** — directory names and known
 config filenames, never file content, never an LLM-invented claim.
-Fast first pass; the deeper analyzers (`project-analysis-*`,
-`standards-from-config`, `module-detect-on-the-fly`) remain the tools
-for filling in what this script cannot know.
+This is the fast first pass; the deeper analyzers
+(`project-analysis-*`, `standards-from-config`, `module-detect-on-the-fly`)
+remain the tools for filling in what this script cannot know.
 
 ## Steps
 
@@ -81,8 +81,9 @@ Then:
 ./scripts-run src/scripts/generate_knowledge_index
 ```
 
-The team-sharing gate still applies at commit time — none of these
-pages should carry `visibility: private`.
+The [team-sharing gate](../../../scripts/check_knowledge_sharing.ts)
+still applies at commit time — none of these pages should carry
+`visibility: private`.
 
 ### 4. Discard the rest
 
@@ -100,6 +101,6 @@ gitignored and regenerates on the next bootstrap run).
 
 ## See also
 
-- [`/team-knowledge consolidate`](consolidate.md) — the ongoing capture loop this one-shot seed feeds into.
-- [`standards-from-config`](../../skills/standards-from-config/SKILL.md) — the full pointer+digest derivation for coding standards.
-- [`project-analysis-core`](../../skills/project-analysis-core/SKILL.md) / [`module-detect-on-the-fly`](../../skills/module-detect-on-the-fly/SKILL.md) — the deeper analyzers for `modules.md`.
+- [`/team-knowledge consolidate`](../consolidate/command.md) — the ongoing capture loop this one-shot seed feeds into.
+- [`standards-from-config`](../../../skills/standards-from-config/SKILL.md) — the full pointer+digest derivation for coding standards.
+- [`project-analysis-core`](../../../skills/project-analysis-core/SKILL.md) / [`module-detect-on-the-fly`](../../../skills/module-detect-on-the-fly/SKILL.md) — the deeper analyzers for `modules.md`.

package/dist/agent-src/commands/team-knowledge/consolidate.md

@@ -24,7 +24,7 @@ Turns the gitignored, in-flight typed-observation events
 `context_stale` — see `src/scripts/_lib/knowledge_events.ts`) into
 committed `agents/knowledge/` pages. This is the ONLY path from intake
 to a tracked page — no skill or rule writes a tracked knowledge page
-mid-task (see [`knowledge-pages`](../../templates/contexts/knowledge-pages.md)
+mid-task (see [`knowledge-pages`](../../../agent-src/templates/contexts/knowledge-pages.md)
 § Storage location).
 
 ## Steps
@@ -44,7 +44,7 @@ No pending topics → tell the user and stop. Nothing to do.
 ### 2. Triage each topic
 
 For each aggregate in the report, apply the same taxonomy
-[`memory-consolidation`](../../skills/memory-consolidation/SKILL.md)
+[`memory-consolidation`](../../../skills/memory-consolidation/SKILL.md)
 uses for engineering-memory facts:
 
 | Triage | Condition | Action |
@@ -61,7 +61,7 @@ events before deciding.
 ### 3. Draft and present the batch
 
 For every `NEW` / `EXTEND` / `CONFLICT` aggregate, draft the page
-content (following [`knowledge-pages`](../../templates/contexts/knowledge-pages.md) —
+content (following [`knowledge-pages`](../../../agent-src/templates/contexts/knowledge-pages.md) —
 `date + what + why` for session entries, ≤ 200 lines per page) and
 present the full batch to the user:
 
@@ -81,7 +81,7 @@ present the full batch to the user:
 
 On approval:
 
-1. Write each approved page (respecting the team-sharing gate —
+1. Write each approved page (respecting the [team-sharing gate](../../../scripts/check_knowledge_sharing.ts) —
    never set `visibility: private` on a page landing here).
 2. Regenerate the index: `./scripts-run src/scripts/generate_knowledge_index`.
 3. Clear the consumed intake: `./scripts-run src/scripts/consolidate_knowledge_events --commit`
@@ -96,9 +96,10 @@ On approval:
   land in the page, human resolves later.
 - Never writes a page with `visibility: private` (belongs in the
   ADR-100 global store instead).
-- Do NOT commit the written pages — that is the user's call.
+- Do NOT commit the written pages — that is the user's call, per
+  [`commit-policy`](../../../rules/commit-policy.md).
 
 ## See also
 
-- [`memory-consolidation`](../../skills/memory-consolidation/SKILL.md) — the sibling loop for the 5 fixed engineering-memory types; this command is its knowledge-page counterpart.
-- [`learning-to-rule-or-skill`](../../skills/learning-to-rule-or-skill/SKILL.md) — where a page's recurring topic graduates into a skill/rule once it crosses the skill-candidate threshold.
+- [`memory-consolidation`](../../../skills/memory-consolidation/SKILL.md) — the sibling loop for the 5 fixed engineering-memory types; this command is its knowledge-page counterpart.
+- [`learning-to-rule-or-skill`](../../../skills/learning-to-rule-or-skill/SKILL.md) — where a page's recurring topic graduates into a skill/rule once it crosses the skill-candidate threshold.

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

@@ -77,7 +77,10 @@ a wrapper is invoked directly.
 
 ## Sub-commands
 
-| Sub-command | Delegates to (Claude Code) | Purpose |
+On Claude Code each sub-command **delegates** to the paired `/codex:*` command
+rather than routing to an in-suite skill — the second column names that target.
+
+| Sub-command | Routes to | Purpose |
 |---|---|---|
 | `/team review [--background]` | `/codex:review` | Cross-model review of the current working-tree / staged diff |
 | `/team adversarial <focus>` | `/codex:adversarial-review` | Adversarial cross-model review on a named focus — the escalation rung above the single-model `adversarial-review` skill |

package/dist/agent-src/commands/upstream-contribute.md

@@ -131,7 +131,6 @@ Frontmatter must have `source: package`.
 ./scripts-run src/scripts/check_condensation            # No 🔴 errors for this file
 ./scripts-run src/scripts/condense --generate-tools    # Regenerate symlinks
 bash scripts/condense.sh --check                # dist/agent-src/ in sync with source
-bash scripts/condense.sh --check-hashes         # All hashes match
 ```
 
 Fix any issues before continuing.

package/dist/agent-src/contexts/authority/commit-mechanics.md

@@ -99,3 +99,10 @@ any roadmap-producing flow), do **not** include commit steps unless
 the user explicitly requested them — commits are a delivery decision;
 roadmaps plan **work**. If the user explicitly wants commit steps,
 write them clearly (e.g. "Commit phase X: chore: …").
+
+## Migrated from `commit-policy` (kernel-budget relief, 2026-07-30)
+
+### See also
+
+- [`scope-control`](../../../rules/scope-control.md) — git-ops permission gate (push, merge, branch, PR, tag).
+- [`no-cheap-questions`](../../../rules/no-cheap-questions.md) — canonical Iron Law. · [`autonomous-execution`](../../../rules/autonomous-execution.md) · [`/commit`](../../commands/commit.md) · [`/commit:in-chunks`](../../commands/commit/in-chunks.md).

package/dist/agent-src/contexts/authority/destructive-mechanics.md

@@ -88,3 +88,19 @@ safe to remove or replace.
 - Reading a roadmap step listing files to delete as authorization to
   *commit* the deletion. The step authorizes the *edit*; the commit
   is row 6 of the Hard Floor and needs its own confirmation.
+
+## Migrated from `non-destructive-by-default` (kernel-budget relief, 2026-07-30)
+
+### Enforcement — stated honestly (`enforced_by: none`)
+
+This floor has **no mechanical backstop**, and pretending otherwise would be the
+failure it exists to prevent. No script can enforce "ask before you deploy" — the
+obligation is to pause and surface, and the only thing that observes it is the
+human in the loop. Adjacent hooks (`block-no-verify`) guard hook-bypass, not
+destructive intent; claiming them here would inflate coverage, so they are not
+claimed.
+
+So this rule is load-bearing **and** unenforced. It is carried by the model
+honouring it — which is why it is a kernel rule, why it is short, and why it is
+counted as uncovered in the enforcement-coverage report rather than excluded
+from it.

package/dist/agent-src/contexts/authority/scope-mechanics.md

@@ -195,3 +195,64 @@ carry over. Concretely: *"arbeite eigenständig"* given for
 "Roadmap A" is consumed when Roadmap A's deliverable lands. A new
 named task (Roadmap B, ticket Y, "next slice") needs fresh
 authorization. See [`autonomous-execution § task-scope`](../../rules/autonomous-execution.md#task-scope--autonomy-is-bound-to-the-named-task).
+
+## Post-artifact hard stop — no auto-offer of execution
+
+```
+ARTIFACT LANDED (ROADMAP / ADR / PLAN / PITCH / TICKET / DESIGN DOC)
+→ STOP. NEVER ASK "READY TO START?" / "BEGIN PHASE 1?" / "SOFORT
+AUSFÜHREN?" / "SOLL ICH MIT E1.1 ANFANGEN?". EXECUTION STARTS ON A
+LATER TURN WITH AN EXPLICIT EXECUTION VERB FROM THE USER.
+```
+
+Authoring verbs (`create`, `draft`, `write`, `erstelle`, …) authorize
+the artifact only. The moment the file is saved, the turn ends —
+silence on execution. The artifact is the deliverable; the work it
+plans is a **separate decision** the user pins on a later turn.
+
+This anchor binds three command surfaces:
+
+- [`/roadmap:create` Step 9](../../commands/roadmap/create.md) —
+  hand-back is a hard stop; never propose `/roadmap:process-*` after
+  save.
+- [`/challenge-me vision` Step 6 (`!roadmap` branch)](../../commands/challenge-me/vision.md) —
+  routing into `/roadmap:create` ends at file save; the chain does
+  **not** chain into execution.
+- `/adr:create`, `/feature-roadmap`, `/feature-plan`, any future
+  authoring command — same rule, no per-command opt-out.
+
+**Anti-pattern catalogue** (each of these IS the violation, not a
+near-miss):
+
+- *"Roadmap fertig. Soll ich mit Phase A anfangen?"* — execution
+  offer disguised as a status line.
+- Numbered options `1. Start now / 2. Review first / 3. Cancel`
+  after the save line — the user did not ask for an execution menu.
+- *"Recommended next step: `/roadmap:process-phase`"* in the
+  hand-back — back-channels the same ask.
+- Treating *"sieht gut aus"* / *"passt"* / *"alright"* / thumbs-up
+  as authorization to start. None of those are execution verbs.
+- Soft-routing prose: *"the roadmap is ready, kicking off E1.1…"* —
+  hand-off to execution disguised as continuation.
+- Council-review hop counted as an execution authorization: review
+  ends, agent jumps straight to Phase 1.
+
+**Allowed**
+
+- A single hand-back line citing the saved path and any regenerated
+  dashboard count.
+- A "Recommended next step" line in the artifact body itself (e.g.
+  in a pitch) that **names a command** the user can copy — that is
+  reference material in the artifact, not a routing offer in the
+  reply.
+- Continuing on the **same turn** if and only if the user already
+  used an explicit execution verb alongside the authoring verb
+  (*"create the roadmap and start Phase A"*). Mixed-verb handling:
+  default to authoring-only and ask; see [`Authoring vs.
+  implementation`](#authoring-vs-implementation--verb-discipline).
+
+**Bypass** — the user resumes with an explicit execution verb on a
+later turn: `implement`, `build`, `start Phase A`, `arbeite die
+Roadmap ab`, `mach weiter mit Phase 1`. That verb — not the prior
+authoring turn, not a council pass, not a thumbs-up — lifts the
+post-artifact stop.

package/dist/agent-src/contexts/communication/rules-auto/reply-close-mechanics.md

@@ -1,43 +1,45 @@
 # Reply close — work summary + PR link
 
 Lookup material for the reply-close carve-out in the
-[`direct-answers`](../../../rules/direct-answers.md) Brevity Iron Law. Rule body
-states the obligation in one line; this file is the detail.
+[`direct-answers`](../../../rules/direct-answers.md) Brevity Iron Law. The rule
+body states the obligation in one line; this file is the detail.
 
 ## The carve-out
 
 Brevity governs the reply **body**; the **close** is the exception.
 
-- Reply that landed **substantial work** — multi-step change, roadmap/branch
+- A reply that landed **substantial work** — a multi-step change, roadmap/branch
   progress, or a created/updated PR — ends with **ONE** compact status summary at
   the very end, never mid-reply: what's done, what remains.
-- **Simple one-answer replies still skip the summary** (Brevity Iron Law unchanged
-  for them). Summary is for work replies, not every turn.
+- **Simple one-answer replies still skip the summary** (the Brevity Iron Law is
+  unchanged for them). The summary is for work replies, not every turn.
 
 ## PR link is the literal last line
 
-PR **created or updated this turn** → its raw GitHub URL is the **literal last
-line**, after the summary, nothing below it — so the user never hunts for it. One
-PR → one URL line.
+If a PR was **created or updated this turn**, its raw GitHub URL is the **literal
+last line** of the reply — after the summary, nothing below it — so the user never
+hunts for it. One PR → one URL line.
 
 - Only when a PR was actually created/updated **this turn**. No PR → no link line.
-- Raw URL (not a masked markdown link) so it is unmistakable and clickable.
+- The raw URL (not a masked markdown link) so it is unmistakable and clickable.
 
 ## Why it does not collide with sibling rules
 
-- **`language-and-tone`** last-line rule: a raw URL is language-neutral → satisfies
-  the "every user-visible token mirrors the user's language" gate.
+- **`language-and-tone`** last-line rule: a raw URL is language-neutral, so it
+  satisfies the "every user-visible token mirrors the user's language" gate.
 - **`no-pr-progress-comments`** / **`no-attribution-footers`**: those govern
-  comments posted *on the PR*. Summary + URL here are the **chat reply body**, not
-  a PR comment — neither applies.
-- **`user-interaction`**: a numbered-options block, when present, still comes last
-  among interactive elements; the PR URL sits below it as the final line and is not
-  an "open question" → does not trip the no-trailing-open-question rule.
+  comments posted *on the PR*. The summary and URL here are the **chat reply
+  body**, not a PR comment — so neither rule applies.
+- **`user-interaction`**: a numbered-options block, when present, still comes
+  last among interactive elements; the PR URL sits below it as the final line and
+  is not an "open question," so it does not trip the no-trailing-open-question
+  rule.
 
 ## Failure modes
 
 - Burying the summary mid-reply, then continuing with more detail below it.
-- Emitting the summary for a trivial one-answer reply (brevity says skip).
-- Creating a PR but leaving its URL mid-reply, forcing the user to scroll/search.
+- Emitting the summary for a trivial one-answer reply (brevity says skip it).
+- Creating a PR but leaving its URL somewhere in the middle, forcing the user to
+  scroll/search.
 - Masking the URL behind link text so it is not obviously the PR.
 - Adding a PR-link line when no PR was created this turn.

package/dist/agent-src/contexts/communication/rules-auto/skill-quality-mechanics.md

@@ -25,7 +25,7 @@ Make descriptions "pushy" — explicit about when to fire:
   adjectives, drop the second example phrasing, or collapse a list — do
   **not** drop the trigger vocabulary or the `even if ...` tail.
 
-Source: [`skills/skill-creator` in `an external reference`](https://github.com/an external reference/blob/main/skills/skill-creator/SKILL.md).
+Source: adapted from an external reference.
 
 **Litmus test:** Read the description cold, without the skill's body. If you
 cannot name at least two phrasings a user would realistically type that should

package/dist/agent-src/contexts/communication/rules-auto/slash-command-routing-policy-mechanics.md

@@ -4,7 +4,7 @@ Lookup table for the `slash-command-routing-policy` rule. Lists the
 locked clusters and their sub-commands so the rule itself can stay at
 its current LOC while still reflecting the full surface. Source of
 truth for the cluster names is
-[`docs/contracts/command-clusters.md`](../../../../../../../docs/contracts/command-clusters.md);
+[`docs/contracts/command-clusters.md`](../../../../../docs/contracts/command-clusters.md);
 this file mirrors that contract for runtime lookup. Linter:
 `scripts/check_cluster_patterns.ts` (verifies dispatcher shape).
 

package/dist/agent-src/contexts/communication/rules-auto/source-of-truth-mechanics.md

@@ -12,17 +12,17 @@ when authoring or pre-review verification fires.
 1. **Create or edit** the file in `src/{path}` (e.g. `src/rules/`, `src/skills/`,
    `src/agent-src/`, `src/domains/<pack>/`)
 2. **Do NOT auto-condense.** Continue working.
-3. **Before commit/push:** Check if condensation is needed
-   (`bash src/scripts/condense.sh --changed`). If files need condensation,
-   ask the user:
+3. **Before commit/push:** Check whether any projection is out of date
+   (`bash src/scripts/condense.sh --changed` — `dist != rewrite(src)`). If files
+   are listed, ask the user:
    ```
-   > 📦 {N} dist/agent-src files need condensation before commit.
+   > 📦 {N} dist/agent-src files are out of date before commit.
    >
-   > 1. Condense now — run /condense
-   > 2. Later — commit without condensation
+   > 1. Sync now — run /condense
+   > 2. Later — commit without syncing
    ```
-4. If condensing: run `/condense` command, then
-   `bash src/scripts/condense.sh --mark-done {path}`
+4. If syncing: run the `/condense` command — `--sync` writes the projection,
+   nothing has to be marked afterwards.
 
 For new non-.md files (`.php`, configs):
 `bash src/scripts/condense.sh --sync` copies them automatically.
@@ -94,6 +94,5 @@ Commands have `disable-model-invocation: true` in their frontmatter.
 | Create new non-`.md` | Create in `src/`, run `bash src/scripts/condense.sh --sync` |
 | Create new command | Create in `src/agent-src/commands/`, sync, `./scripts-run src/scripts/condense --generate-tools` |
 | Delete a file | Delete from `src/` and `dist/agent-src/` |
-| Check what needs condensation | `bash src/scripts/condense.sh --changed` |
-| Mark file as condensed | `bash src/scripts/condense.sh --mark-done {path}` |
+| Check which projections are out of date | `bash src/scripts/condense.sh --changed` |
 | Verify everything is in sync | `bash src/scripts/condense.sh --check` |

package/dist/agent-src/contexts/communication/rules-auto/think-before-action-mechanics.md

@@ -43,7 +43,7 @@ behavior. Cannot → read more.
 ## Consult memory before editing
 
 Invariants live in the memory layer; architectural rationale lives in ADRs.
-Via [`memory-access`](../../../../../../docs/guidelines/agent-infra/memory-access.md),
+Via [`memory-access`](../../../../../docs/guidelines/agent-infra/memory-access.md),
 call `retrieve(types=["domain-invariants"], keys=<touched paths>, limit=3)` —
 a matching `domain-invariant` is a hard constraint. For *why* the current
 shape exists, check the ADR index `docs/decisions/INDEX.md`. Cite the `id` /

package/dist/agent-src/contexts/communication/rules-auto/user-interaction-mechanics.md

@@ -180,6 +180,17 @@ When processing multiple items (e.g., review comments, test failures), show prog
 **Comment 3/7** — `filename.php:42`
 ```
 
+## Bullet floor
+
+Multi-step content earns bullets; each bullet is a complete 1–2-sentence
+statement, never a fragment masquerading as a list item ("Faster." /
+"Better UX." with no subject or verb). A decline or refusal is **never**
+rendered as a bulleted list of reasons — declines are short prose (one or
+two sentences), because a bulleted refusal reads as a checklist to argue
+against rather than a direct answer.
+
+Full example set: [`asking-and-brevity-examples § Bullet floor`](../../../../../../docs/guidelines/agent-infra/asking-and-brevity-examples.md#bullet-floor).
+
 ## Summaries
 
 After completing a batch of actions, provide a summary table:

package/dist/agent-src/contexts/contracts/artifact-engagement-flow.md

@@ -211,26 +211,37 @@ $ grep agent-engagement .gitignore
 The fastest path is the project's own validator — it enforces every
 rule listed above:
 
+```bash
+# Validate every line against the schema + redaction floor
 ```bash
 # Validate every line against the schema + redaction floor
 node -e '
-import pathlib, sys
-sys.path.insert(0, ".agent-src.uncondensed/templates/scripts")
-from telemetry.engagement import EngagementSchemaError, parse_event
-log = pathlib.Path(".agent-engagement.jsonl")
-ok = bad = 0
-for i, line in enumerate(log.read_text().splitlines(), 1):
-    if not line.strip():
-        continue
-    try:
-        parse_event(line + "\n")
-        ok += 1
-    except EngagementSchemaError as e:
-        bad += 1
-        print(f"line {i}: {e}", file=sys.stderr)
-print(f"{ok} valid, {bad} rejected")
+const fs = require("fs");
+const lines = fs.readFileSync(".agent-engagement.jsonl", "utf8").split("
+").filter(Boolean);
+let ok = 0, bad = 0;
+for (const [i, line] of lines.entries()) {
+    try { JSON.parse(line); ok++; }
+    catch (e) { bad++; console.error(`line ${i+1}: ${e.message}`); }
+}
+console.log(`${ok} valid, ${bad} rejected`);
 '
 
+# A bad-line spot-check that does not depend on the validator
+node -e '
+const fs = require("fs");
+const forbidden = /[/\]|\.[a-z0-9]{1,10}$/i;
+for (const line of fs.readFileSync(".agent-engagement.jsonl", "utf8").split("
+").filter(Boolean)) {
+    const obj = JSON.parse(line);
+    for (const kind of ["consulted", "applied"])
+        for (const ids of Object.values(obj[kind] ?? {}))
+            for (const v of ids)
+                if (forbidden.test(v) || v.length > 200) console.log("LEAK:", v);
+}
+'
+```
+
 # A bad-line spot-check that does not depend on the validator
 node -e '
 import json, pathlib, re

package/dist/agent-src/contexts/contracts/command-suggestion-flow.md

@@ -100,6 +100,9 @@ journey that self-suggests; the sub-command is reached by routing from the head
 always works. This keeps the proactive surface to a few journeys, not one
 entry point per mode (9.4.0 review; `road-to-surface-consolidation`, council
 2026-07-20). Standalone commands (no `sub:`) and cluster heads may be eligible.
+Rationale: 160 eligible commands all competing to self-suggest IS the "190
+gleichwertig sichtbare Einstiegspunkte" the review flags; the sub-command flip
+collapses it without removing any command.
 
 ## Subordination — when to stay silent
 

package/dist/agent-src/contexts/execution/auto-dispatch-classification.md

@@ -21,20 +21,20 @@ A task is classified **delegable** when **any** of these holds:
    `parallelizable: steps | files | independent` in its frontmatter.
    - `steps` → ordered plan → `do-in-steps`.
    - `files` / `independent` → independent slices → `do-in-parallel`.
-2. **Ordered-plan structure** — explicit ordered plan (numbered steps / a roadmap
-   phase / a checklist) → `do-in-steps`. Deterministic markers:
-   - `1. … 2. … 3.` numbered list in the user message.
-   - References a roadmap phase or checklist.
-   - "first … then … finally" with distinct deliverables.
-   - "in N steps" or "phase by phase".
+2. **Ordered-plan structure** — the task is an explicit ordered plan (numbered
+   steps / a roadmap phase / a checklist) → `do-in-steps`. Deterministic markers:
+   - User message contains `1. … 2. … 3.` numbered list.
+   - User message references a roadmap phase or checklist.
+   - User message uses "first … then … finally" with distinct deliverables.
+   - User message says "in N steps" or "phase by phase".
 3. **Independent-slices structure** — N ≥ 3 independent targets of the same shape
    → `do-in-parallel`. Deterministic markers (N ≥ 3 of the same form):
-   - N file paths, each a separate analysis/edit target.
-   - N named modules/components for the same action.
+   - N file paths listed where each is a separate analysis/edit target.
+   - N named modules/components to perform the same action on.
    - N named test files, adapters, endpoints, or services to convert/review/audit.
-   - "for each X in [list]" where list ≥ 3 items, no cross-item dependency.
-   Do **not** fire on interdependent lists (e.g. "add these 3 sequential
-   migrations") — those are ordered plans (signal 2).
+   - "for each X in [list]" where list has ≥ 3 items and no cross-item dependency.
+   Do **not** fire on lists where items are interdependent (e.g. "add these 3
+   sequential migrations") — those are ordered plans (signal 2).
 
 AND the **task-size floor** is cleared: the task's estimated size exceeds a
 minimum (trivial one-line edits never delegate — the dispatch overhead
@@ -58,9 +58,10 @@ dwarfs the work). Below the floor → in-session.
 ## Per-slice tier inference (v1.5 — deterministic, task-TYPE-keyed)
 
 Once a slice is classified **delegable**, a second deterministic table infers
-its `model_tier` (road-to-cost-aware-model-routing, council 2026-07-08).
-Keyed **exclusively on the classifier's task-TYPE outputs** — never raw size
-metrics; diff size anti-correlates with difficulty in refactoring domains.
+its `model_tier` (road-to-cost-aware-model-routing, council 2026-07-08). The
+inference is keyed **exclusively on the classifier's task-TYPE outputs** —
+never on raw size metrics; diff size anti-correlates with difficulty in
+refactoring domains.
 
 ```
 UNKNOWN / AMBIGUOUS → inherit (SESSION TIER). NEVER GUESS DOWN.
@@ -76,14 +77,42 @@ THEY NEVER CREATE ONE.
 | Delegable + synthesis / judgment (review, analysis slice) | `medium` — judge one tier up per the orchestration Iron Law |
 | Any other / ambiguous shape | `inherit` — session tier, no downshift |
 
-**Negative size guard:** slice scope exceeding the mechanical envelope
-(multi-file mutation, diff surface beyond single responsibility) loses `lite`
-candidacy, resolves one row down — size never argues FOR a downshift, only
-against one.
+**Negative size guard:** a slice whose scope exceeds the mechanical envelope
+(multi-file mutation, or a diff surface beyond a single responsibility) loses
+its `lite` candidacy and resolves one row down — size never argues FOR a
+downshift, only against one.
 
-Every inferred decision records `tier_source: "inferred"` in orchestration
-telemetry (static pins record `"static"`; session-tier runs `"inherit"`) so
-the evidence gate scores inferred routing separately from static pinning.
+Every inferred decision records `tier_source: "inferred"` in the
+orchestration telemetry (statically pinned tiers record `"static"`;
+session-tier runs record `"inherit"`) so the evidence gate can score inferred
+routing separately from static pinning.
+
+## Dispatch-primitive output — fork vs. named subagent (road-to-cache-economy Phase 4)
+
+A third, ORTHOGONAL classifier output — independent of form and tier — on a
+host that exposes a fork primitive alongside named-subagent dispatch:
+
+```
+TOOL AND SCOPE FIT IS FIRST-ORDER. CACHE INHERITANCE IS SECOND-ORDER.
+NEVER "ALWAYS FORK" — THE BENEFIT CANNOT BE PREDICTED BEFORE THE FORK HAPPENS.
+```
+
+| Slice shape | `dispatch_primitive` |
+|---|---|
+| Continues the parent's task under IDENTICAL tools and constraints (no isolation need, no nested dispatch) | `fork` — its first request reads the parent's cache |
+| Needs isolation, a different tool set, or nested dispatch | `subagent` — a fork cannot nest and cannot change tools |
+| Ambiguous / host has no fork primitive | `subagent` — the only primitive this suite's `Agent`-tool dispatch uses today |
+
+A fork forces background mode, which changes the tool set and therefore
+invalidates the very prefix that motivated the fork — so a slice that would
+need a different tool set mid-flight is a `subagent` candidate even when it
+started as a fork candidate. This output never downgrades the form or tier
+decisions above; it only picks the primitive once a `parallel` / `steps` /
+`judge` form has already been selected. See
+[`subagent-orchestration/SKILL.md`](../../skills/subagent-orchestration/SKILL.md)
+§ Form gate and
+[`prompts/README.md`](../../skills/subagent-orchestration/prompts/README.md)
+§ Prompt-cache discipline for the full rationale.
 
 ## Lookup-class rung (L0 — road-to-lean-agent-init, BELOW the tiers)
 
@@ -100,7 +129,7 @@ NEVER A SILENTLY DEGRADED ANSWER. NO LLM CLASSIFIER FALLBACK (CUT C3).
 
 | Task pattern | Lookup class | Primitive |
 |---|---|---|
-| "where is X defined" / "confirm X's definition location" | `definition` | capped `rg` (definition-shaped patterns, `--max-count`); `code_graph query` only as opportunistic accelerant when an index is present + fresh + `hooks.code_graph.enabled: true` |
+| "where is X defined" / "confirm X's definition location" | `definition` | capped `rg` (definition-shaped patterns, `--max-count`); `code_graph query` only as an opportunistic accelerant when an index is present + fresh + `hooks.code_graph.enabled: true` |
 | "who calls / imports X" / "confirm call sites" | `references` | capped `rg` (reference-shaped patterns); same optional `code_graph affected` accelerant clause |
 | "does string Y exist" / "probe candidate strings" | `string-existence` | FTS one-shot (`memory_lookup` for the knowledge corpus) or capped `rg -n --max-count` for the codebase |
 | "run report Z" / "run check_*" | `report-run` | direct script run, wrapped per the **measured** rtk allowlist (`internal/bench/rtk-savings/RESULTS.md` — wrap only the ~55%-savings class) |
@@ -112,21 +141,22 @@ recall 0.365 vs grep 0.797 on graph-shaped questions, root cause an **indexing
 gap** (TS arrow-function exports produce no symbol nodes) that hits structured
 lookups exactly as it hits NL retrieval — and the recorded consequence bound
 keeps `code_graph.enabled` false permanently. The accelerant clause is the
-escape hatch that bound names ("unless external evidence appears"); today
-inert.
-
-**Escalation, not degradation:** a primitive returning nothing (index miss,
-ambiguous pattern, absent report script) — **or an unusable result** (empty,
-ambiguous between candidates, or visibly off-target because the lookup
-pattern matched a task that was not actually a lookup, e.g. a dynamically
-constructed symbol name) — escalates to the regular classification path —
-answer-quality bar never drops. "Primitive ran but result unusable"
-escalates exactly like "class unknown". `unknown` resolves to the regular
-path (`inherit` semantics), never down-guessed.
-
-Correctness floor: golden comparison in `internal/bench/lean-init/`
-(primitive answer ≡ agent answer on ≥10 lookup tasks, incl. the four observed
-shapes); any mismatch is a routing bug, not a rounding error.
+escape hatch that bound names ("unless external evidence appears"); today it
+is inert.
+
+**Escalation, not degradation:** a primitive that returns nothing (index miss,
+pattern too ambiguous, report script absent) — **or an unusable result**
+(empty, ambiguous between candidates, or visibly off-target because the
+lookup pattern matched a task that was not actually a lookup, e.g. a
+dynamically constructed symbol name) — escalates to the regular
+classification path above — the answer quality bar never drops. The
+"primitive ran but the result is unusable" case escalates exactly like the
+"class unknown" case. `unknown` resolves to the regular path (`inherit`
+semantics), never down-guessed.
+
+Correctness floor: the golden comparison in `internal/bench/lean-init/`
+(primitive answer ≡ agent answer on ≥10 lookup tasks, including the four
+observed shapes); any mismatch is a routing bug, not a rounding error.
 
 ## v2+ (deferred, gated on Phase 6 evidence)
 

package/dist/agent-src/contexts/execution/auto-orchestration-activation.md

@@ -12,7 +12,7 @@ whether dispatch is physically possible on the host).
 | Key | Default | Purpose |
 |---|---|---|
 | `subagents.enabled` | `true` | Global master switch. `false` = the whole layer is off — the canonical kill-switch. No auto-dispatch, no routing, everything runs in-session. |
-| `subagents.auto` | `on` | Auto-dispatch mode: `off` \| `ask` \| `on`. Default `on` on subagent-capable hosts (flipped from `ask` 2026-07-09, ADR-117), `off` elsewhere. |
+| `subagents.auto` | `on` | Auto-dispatch mode: `off` \| `ask` \| `on`. Shipped default `on` on subagent-capable hosts (flipped from `ask` 2026-07-09, ADR-117), `off` elsewhere. |
 | `subagents.downshift` | `true` | Route delegable sub-tasks to the lowest-capable tier (Phase 2). |
 | `subagents.quota_arbitrage` | `true` | Prefer a separate-quota-pool model where the host manifest allows it (Phase 2 bonus). |
 | `subagents.model_map` | `{}` | Optional per-tier model override for downshift routing. |

package/dist/agent-src/contexts/execution/autonomy-examples.md

@@ -53,7 +53,7 @@ or `auto`-after-opt-in, act on them.
 - "Should we commit now or after the next change?" — answered by the
   commit-default in [`commit-policy`](../../rules/commit-policy.md), no need to ask.
 - "How should I split the commits?" — never asked; either you are
-  invoked via `/commit-in-chunks` (split and commit) or you are not
+  invoked via `/commit:in-chunks` (split and commit) or you are not
   (don't commit at all).
 - "Should I run the linter / tests now or after the change?" — run
   what [`verify-before-complete`](../../rules/verify-before-complete.md)

package/dist/agent-src/contexts/execution/autonomy-mechanics.md

@@ -129,12 +129,13 @@ knob. The per-dimension uncertainty score (see
 [`notes-first-reasoning`](../../rules/notes-first-reasoning.md)) feeds this
 decision. Engage per [`rdp-gate`](rdp-gate.md).
 
-**Concrete diminishing-returns stop-signal.** Operationalize "marginal evidence
-drops" so it is checkable, not vibes: when **two consecutive investigative probes**
-(a read, a grep, a query) yield **no new load-bearing information** — nothing that
-changes the plan or the hypothesis — stop gathering and act on what you have (or
-ask). The diminishing-returns twin of N=3: N=3 stops fruitless *fixing*, this stops
-fruitless *searching*. Never overrides the read-loop abort in
+**Concrete diminishing-returns stop-signal.** "Marginal evidence drops" is
+otherwise fuzzy; operationalize it so it is checkable, not vibes: when **two
+consecutive investigative probes** (a read, a grep, a query) yield **no new
+load-bearing information** — nothing that changes the plan or the hypothesis —
+stop gathering and act on what you have (or ask). This is the diminishing-returns
+twin of the N=3 hard cap: N=3 stops fruitless *fixing*, this stops fruitless
+*searching*. It never overrides the read-loop abort in
 [`context-hygiene`](../../rules/context-hygiene.md) — that is the hard ceiling;
 this is the earlier soft signal.