@event4u/agent-config 9.28.0 → 9.29.0

package/dist/agent-src/skills/forensics-report/SKILL.md

@@ -0,0 +1,114 @@
+---
+name: forensics-report
+description: "Use when a release review needs machine-derived evidence from git history — hotspot risk and change-coupling analyzers. Triggers on 'hotspot', 'what changes together', 'release forensics'."
+domain: engineering
+model_tier: medium
+workspaces:
+  - engineering
+packs:
+  - forensics
+lifecycle: experimental
+trust:
+  level: experimental
+install:
+  default: false
+  removable: true
+---
+
+# forensics-report
+
+## When to use
+
+- Before or during a **release review**, when machine-derived evidence is
+  wanted alongside the human review: "which files are the risk hotspots in
+  this release", "what changes together", "release forensics".
+- When intended module boundaries should be checked against what the git
+  history actually shows (change-coupling across module lines).
+- The output is **advisory, never a gate** — it points reviewers at places
+  worth attention; it never passes or fails anything.
+- NOT for IT-security incident forensics (route to `incident-commander`) and
+  NOT for runtime performance hotspots (route to `performance-analysis`).
+
+## Procedure
+
+1. **Pick the range** — the previous release tag to HEAD (`vPREV..HEAD`) or a
+   released pair (`vPREV..vNEW`). Pin the range explicitly; a floating range
+   breaks determinism.
+2. **Run the report:**
+
+   ```bash
+   npx tsx node_modules/@event4u/agent-config/src/scripts/forensics_report.ts --range vPREV..vNEW --out report.json
+   ```
+
+   Two deterministic, git-log-based analyzers run, read-only and offline:
+   - **Hotspot risk** = normalized change frequency x normalized complexity
+     per file over the range.
+   - **Change-coupling** = `co_changes(A,B) / min(changes(A), changes(B))`
+     per file pair.
+3. **Optionally feed the release findings ledger:** add
+   `--findings-out findings.json`, then ingest via
+
+   ```bash
+   npx tsx node_modules/@event4u/agent-config/src/scripts/check_finding_dispositions.ts --ingest findings.json --release X.Y.Z
+   ```
+
+   which appends to the existing ledger at
+   `agents/evidence/release-findings/<version>.json`. Emitted findings are
+   `kind: correctness` with `severity: low|medium` — non-blocking for the
+   release gate by construction.
+4. **Read the results:**
+   - `hotspots[]` — a hotspot is a place for reviewer attention, NEVER a
+     quality verdict. A frequently changed complex file can be exactly right
+     as it is; the score says "look here", not "this is bad".
+   - `coupling[]` — file pairs that habitually change together.
+   - `boundary_contradictions[]` — cross-module coupling above the threshold,
+     i.e. a contradiction between intended and actual boundaries. This is a
+     finding CLASS to discuss, not a failure.
+
+## Output format
+
+1. Lead with the scanned denominator (`scanned.commits` / `scanned.files`, including skipped bulk commits) — a reader must be able to judge the sample before the findings.
+2. Present hotspots and coupling as ranked tables with the score interpretation line ("a place to look, not a quality verdict") attached.
+3. When `--findings-out` was used, state the ingest command and the ledger path the findings landed in.
+
+Deterministic JSON with these fields:
+
+| Field | Meaning |
+|---|---|
+| `mode` | analyzer mode the run used |
+| `range` | the pinned rev-range analyzed |
+| `scanned` | commit/file counts, including skipped bulk commits |
+| `params` | thresholds and normalization parameters in effect |
+| `hotspots[]` | per-file hotspot entries (path, frequency, complexity, score) |
+| `coupling[]` | file-pair coupling entries (paths, co-change ratio) |
+| `boundary_contradictions[]` | cross-module coupling hits above threshold |
+
+Findings emitted via `--findings-out` conform to
+`review-findings.schema.json` (shipped with the package; `kind: correctness`,
+`severity: low|medium`).
+
+## Gotchas
+
+- **Bulk commits are coupling noise** — commits touching more than
+  `max_commit_files` files (default 50) are skipped by the coupling analyzer
+  and counted in the report's `scanned` section.
+- **Only git log + working-tree file metrics are read** — renamed files
+  appear as two separate paths, splitting their history across both.
+- **Deterministic only with a pinned range** — `vPREV..vNEW` reproduces;
+  "since today" or a date-relative range does not.
+
+## Do NOT
+
+- Do NOT build a gate out of this report — the pre-registered value question
+  in `docs/CLAIMS.md` (`forensics-pack-value`) decides promotion or closure;
+  until then the output stays advisory.
+- Do NOT sell hotspot scores as a code-quality proxy — they measure change
+  frequency and complexity, not correctness or craftsmanship.
+- Do NOT emit findings with `severity: high|critical` or
+  `kind: security|claim` — that would block the release gate, which this
+  analyzer must never do.
+
+## See also
+
+- `forensics_report.ts` (this package's scripts tree; consumer path `node_modules/@event4u/agent-config/src/scripts/forensics_report.ts`) — the analyzer implementation.
+- `docs/CLAIMS.md` — the pre-registered `forensics-pack-value` question.

package/dist/agent-src/skills/forensics-report/evals/triggers.json

@@ -0,0 +1,17 @@
+{
+    "skill": "forensics-report",
+    "last_eval": "2026-08-09",
+    "description": "5 should-trigger + 5 should-not-trigger. Should-trigger covers DE+EN phrasings for git-history release forensics (hotspot risk, change-coupling, release-review evidence). Should-not-trigger covers near-miss neighbours: security incident forensics (incident-commander), plain git log, runtime performance hotspots (performance-analysis), line blame, security audit.",
+    "queries": [
+        { "q": "which files are the riskiest hotspots in this release", "trigger": true },
+        { "q": "what files always change together", "trigger": true },
+        { "q": "run the forensics report for the release review", "trigger": true },
+        { "q": "zeig mir hotspot und coupling evidenz für das release", "trigger": true },
+        { "q": "machine-derived evidence for the release review", "trigger": true },
+        { "q": "do a forensic analysis of the security incident from last night", "trigger": false },
+        { "q": "show me the git log for the last week", "trigger": false },
+        { "q": "find the performance hotspots in production at runtime", "trigger": false },
+        { "q": "who wrote this line — run git blame on it", "trigger": false },
+        { "q": "run a security audit on the auth module", "trigger": false }
+    ]
+}

package/dist/agent-src/skills/premortem/SKILL.md

@@ -80,6 +80,31 @@ For each top failure mode (sorted by L × I from `risk-officer`):
    rollout strategy that reduces the likelihood of this failure mode. Keep it
    executable; flag as `accept` if no practical guardrail exists.
 
+### 3b. Compress into the four-part failure register
+
+When the pre-mortem accompanies a plan artefact (a roadmap, an ADR, a design
+doc), compress the analysis into the **four-part failure register** — the
+fixed, comparable shape the roadmap template's optional `## Pre-mortem`
+section consumes:
+
+1. **Three most probable causes of death, ranked** — one paragraph each,
+   naming the concrete mechanism by which the plan most plausibly dies.
+   Mechanisms, not topics: "the gate's two inputs are unmeasurable from this
+   corpus" is a cause; "measurement risk" is not.
+2. **One untested hidden dependency** — the assumption the plan relies on but
+   never tests.
+3. **One modification that makes failure survivable** — a concrete plan change
+   that converts a fatal outcome into a recoverable one.
+4. **One tripwire metric with a horizon** — a measurable signal plus a
+   threshold and deadline that says early that cause #1 is materializing.
+
+The register was validated blind against this repo's own closed nulls
+(3 of 4 registers named the actual failure cause at rank 1 —
+`agents/evidence/analysis/premortem-blind-retro-validation.md`). The hits were
+all failures of *measurability*; the miss was a mechanism that worked but
+added nothing — weight the register toward "can this experiment answer its
+question" and state redundancy risks explicitly.
+
 ### 4. Optional memory write-back
 
 If the analysis surfaces a pattern worth preserving for future decisions:
@@ -96,7 +121,11 @@ If the analysis surfaces a pattern worth preserving for future decisions:
 1. **Ranked prospective failure modes** — L × I ordered list (from `risk-officer`)
 2. **Early-warning signal per failure mode** — concrete, observable indicator
 3. **Preventive guardrails** — one per failure mode (or `accept` with rationale)
-4. **Optional memory candidate** — drafted to `/memory propose` if the pattern
+4. **Four-part failure register** (when accompanying a plan artefact) — three
+   ranked causes of death · one untested hidden dependency · one
+   survivable-failure modification · one tripwire metric with a horizon
+   (step 3b; feeds the roadmap template's optional `## Pre-mortem` section)
+5. **Optional memory candidate** — drafted to `/memory propose` if the pattern
    clears the admission gate (≥ 2 distinct file paths OR ≥ 3 future decisions)
 
 ## Do NOT

package/dist/agent-src/skills/reasoning-orchestrator/SKILL.md

@@ -82,7 +82,8 @@ Each link **delegates** to the artifact that owns it (no duplication):
    ([`complexity-first-planning`](../complexity-first-planning/SKILL.md)); dispatch
    independent subtasks async ([`subagent-orchestration` § RDP](../subagent-orchestration/SKILL.md)),
    auto-triggered by the [`delegation-policy`](../../rules/delegation-policy.md) rule
-   (gates on `subagents.enabled`/`auto` + host manifest).
+   (gates on the emergency incident switch + host manifest — always-on
+   orchestration carries no per-layer on/off setting).
 5. **audit** — check progress against real tool results
    ([`verify-before-complete`](../../rules/verify-before-complete.md)).
 6. **verify** — fresh-context verifier on the structural-complexity gate

package/dist/agent-src/skills/subagent-orchestration/SKILL.md

@@ -56,15 +56,17 @@ parallel subagents **by default** and keep working while they run (async), rathe
 than blocking on each return — intervene only if one goes off track. Engage per
 [`rdp-gate`](../../contexts/execution/rdp-gate.md).
 
-### Settings-gated auto-dispatch
+### Always-on auto-dispatch
 
 "By default" is governed by the [`delegation-policy`](../../rules/delegation-policy.md)
 rule — the single source of the auto-trigger. It gates on the activation context
 ([`auto-orchestration-activation`](../../contexts/execution/auto-orchestration-activation.md)):
-dispatch only when `subagents.enabled`, `subagents.auto != off`, the host
+dispatch only when `emergency.orchestration_halt` is not set, the host
 manifest reports `subagent_spawn: true`, and the task is classified delegable.
-`auto: ask` → ask once; `auto: on` → surface mode + per-subtask tiers in one
-line; any gate failing → in-session no-op. Never lifts a safety floor.
+A matched signal → dispatch, surfacing mode + per-subtask tiers in one line;
+an ambiguous verdict → ask, always; any gate failing → in-session no-op.
+There is no more per-layer on/off setting (always-on orchestration). Never
+lifts a safety floor.
 
 ## Worker-prompt contract
 
@@ -326,7 +328,7 @@ After every auto-dispatched run, write one audit-log-v1 line
 never hand-author the JSON. Line shape, field semantics, recorder
 invocation, and `token_delta` sourcing priority →
 [`orchestration-telemetry` § Emit procedure](../../agent-src/contexts/execution/orchestration-telemetry.md).
-Skip emit when `subagents.enabled: false` or `spawn_count == 0`
+Skip emit when `emergency.orchestration_halt` is set or `spawn_count == 0`
 (in-session run).
 
 ## Gotcha

package/dist/agent-src/skills/subagent-orchestration/schemas/subagent-status.json

@@ -2,15 +2,15 @@
   "$schema": "https://json-schema.org/draft-07/schema#",
   "$id": "https://event4u.dev/agent-config/schemas/subagent-status.json",
   "title": "Subagent Status Envelope",
-  "description": "Required wire format for every implementer or judge subagent return. Loaded by /do-and-judge, /do-in-steps, and the subagent-orchestration skill. Hand-validated by tests/test_subagent_status_schema.py (no jsonschema runtime dep).",
+  "description": "Required wire format for every implementer or judge subagent return. Loaded by /do-and-judge, /do-in-steps, and the subagent-orchestration skill. No jsonschema runtime dep — the authoritative validators are src/scripts/_lib/subagent_response.ts (body) and src/scripts/_lib/subagent_capsule.ts (CHECKPOINT capsule), covered by tests/scripts/_lib_subagent_response.test.ts and tests/scripts/_lib_subagent_capsule.test.ts.",
   "type": "object",
   "required": ["status", "summary"],
   "additionalProperties": false,
   "properties": {
     "status": {
       "type": "string",
-      "enum": ["DONE", "DONE_WITH_CONCERNS", "NEEDS_CONTEXT", "BLOCKED"],
-      "description": "DONE = work shipped, gate green. DONE_WITH_CONCERNS = work shipped but caller must read concerns[]. NEEDS_CONTEXT = paused, blocking_question[] must be answered. BLOCKED = halted, blocking_reason describes why no path forward."
+      "enum": ["DONE", "DONE_WITH_CONCERNS", "NEEDS_CONTEXT", "BLOCKED", "CHECKPOINT"],
+      "description": "DONE = work shipped, gate green. DONE_WITH_CONCERNS = work shipped but caller must read concerns[]. NEEDS_CONTEXT = paused, blocking_question[] must be answered. BLOCKED = halted, blocking_reason describes why no path forward. CHECKPOINT = the worker reached its budget watermark and hands off a state capsule; done/remaining/generation are required."
     },
     "summary": {
       "type": "string",
@@ -40,6 +40,60 @@
     "next_action": {
       "type": "string",
       "description": "What the caller does now. Optional; orchestrator infers from status when omitted."
+    },
+    "generation": {
+      "type": "integer",
+      "minimum": 1,
+      "description": "CHECKPOINT only. Generation index in the recycling chain; the first dispatch is 1."
+    },
+    "done": {
+      "type": "array",
+      "maxItems": 40,
+      "items": {"type": "string", "minLength": 1, "maxLength": 200, "pattern": "^[^\\n\\r]*$"},
+      "description": "CHECKPOINT only. What is finished — ref tokens (file:line, id, path), never bodies."
+    },
+    "remaining": {
+      "type": "array",
+      "maxItems": 40,
+      "items": {"type": "string", "minLength": 1, "maxLength": 240, "pattern": "^[^\\n\\r]*$"},
+      "description": "CHECKPOINT only. What is left, one short line each."
+    },
+    "decisions": {
+      "type": "array",
+      "maxItems": 40,
+      "items": {"type": "string", "minLength": 1, "maxLength": 240, "pattern": "^[^\\n\\r]*$"},
+      "description": "CHECKPOINT only. Choices already made that the successor must not silently re-open."
+    },
+    "open_risks": {
+      "type": "array",
+      "maxItems": 40,
+      "items": {"type": "string", "minLength": 1, "maxLength": 240, "pattern": "^[^\\n\\r]*$"},
+      "description": "CHECKPOINT only. Risks still open at handoff."
+    },
+    "touched_files": {
+      "type": "array",
+      "maxItems": 40,
+      "items": {"type": "string", "minLength": 1, "maxLength": 200, "pattern": "^[^\\n\\r]*$"},
+      "description": "CHECKPOINT only. Paths / file:line this generation touched — never diffs."
+    },
+    "assumptions": {
+      "type": "array",
+      "maxItems": 40,
+      "description": "Stated premises the successor inherits instead of re-deriving. Valid on a CHECKPOINT capsule AND on a completed worker result — unstated assumptions are the first thing compression drops.",
+      "items": {
+        "type": "object",
+        "required": ["statement", "basis", "epistemic_state"],
+        "additionalProperties": false,
+        "properties": {
+          "statement": {"type": "string", "minLength": 1, "maxLength": 240, "pattern": "^[^\\n\\r]*$"},
+          "basis": {"type": "string", "minLength": 1, "maxLength": 200, "pattern": "^[^\\n\\r]*$", "description": "Ref token the premise rests on."},
+          "epistemic_state": {
+            "type": "string",
+            "enum": ["verified", "assumed", "gap"],
+            "description": "The Evidence-Report buckets from evidence-discipline.md — the capsule does not get a private grading scale."
+          }
+        }
+      }
     }
   },
   "allOf": [
@@ -58,6 +112,10 @@
     {
       "if": {"properties": {"status": {"const": "BLOCKED"}}, "required": ["status"]},
       "then": {"required": ["blocking_reason"]}
+    },
+    {
+      "if": {"properties": {"status": {"const": "CHECKPOINT"}}, "required": ["status"]},
+      "then": {"required": ["done", "remaining", "generation"]}
     }
   ]
 }

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

@@ -61,7 +61,7 @@ Generate ≥ 1 concrete case per HIGH dimension, plus the happy path. Enforce th
 
 Widen the list with a second pair of eyes:
 
-- **Subagent path** — when the host supports subagent spawn and subagents are enabled (`subagents.enabled`, `subagents.auto` ≠ off): spawn **one** subagent with an adversarial lens — *"Behavior: {sentence}. Case list so far: {list}. As a security researcher / SRE / chaos engineer: which non-obvious failure mode or abuse vector is missing?"* On security-critical surfaces a **second** subagent with a different lens is allowed; more is waste. Merge only novel cases.
+- **Subagent path** — when the host supports subagent spawn and no emergency halt is set (always-on orchestration: there is no more per-layer on/off setting): spawn **one** subagent with an adversarial lens — *"Behavior: {sentence}. Case list so far: {list}. As a security researcher / SRE / chaos engineer: which non-obvious failure mode or abuse vector is missing?"* On security-critical surfaces a **second** subagent with a different lens is allowed; more is waste. Merge only novel cases.
 - **Degraded path (mandatory when no subagent runs)** — a self-review pass with an explicit lens shift: re-walk step 2 asking *"what breaks in production? what would an attacker try?"* and emit the same output shape (1–2 additional cases, or explicitly "none found").
 
 Unavailability of subagents never skips this step — it only switches the path.

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

@@ -1,6 +1,6 @@
 # {{project_name}}
 
-<!-- Fill placeholders or run `/agents init`; delete after. Capability bullets, not paths. Recipes: `.augment/contexts/contracts/agents-md-anatomy.md`. -->
+<!-- Fill placeholders or run `/agents init`; delete after. Recipes: `.augment/contexts/contracts/agents-md-anatomy.md`. -->
 
 > {{project_description}}
 
@@ -8,8 +8,8 @@
 
 | Layer | Location | Edits |
 |---|---|---|
-| Installed package | `.augment/`, `dist/agent-src/` | hands-off — managed by `event4u/agent-config` |
-| Project layer | `agents/`, `agents/overrides/`, `.agent-settings.yml` | your customizations and config |
+| Installed package | `.augment/`, `dist/agent-src/` | managed by `event4u/agent-config` |
+| Project layer | `agents/`, `agents/overrides/`, `.agent-settings.yml` | your customizations |
 
 ## Pointers
 
@@ -18,7 +18,7 @@
 - **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/).
-- **Delegation & review** — dispatch independent-slice tasks to subagents instead of serial work; a mutating session ends after a neutral cross-model review: [`delegation-policy`](.augment/rules/delegation-policy.md).
+- **Delegation & review** — dispatch independent-slice tasks to subagents instead of serial work; in a team (host-flag gated) the lead coordinates, teammates implement; a mutating session ends after a neutral cross-model review: [`delegation-policy`](.augment/rules/delegation-policy.md).
 
 ## Emergency triage — read this when nothing else is reachable
 

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

@@ -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: "9.28.0"
+agent_config_version: "9.29.0"
 
 # --- Project identity ---
 project:

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

@@ -251,6 +251,20 @@ Templates for roadmap files stored in `agents/roadmaps/` or `{module_root}/{Modu
     `status: draft` roadmaps are exempt until flipped to ready. Enforced
     by `lint_plan_risk_register` at pre-push + CI.
 
+25. **`## Pre-mortem` is optional — and only worth having when it is a
+    forecast.** A roadmap MAY carry a `## Pre-mortem` section holding the
+    four-part failure register from the
+    [`premortem`](../skills/premortem/SKILL.md) skill (three ranked causes
+    of death · one untested hidden dependency · one survivable-failure
+    modification · one tripwire metric with a horizon). It is deliberately
+    NOT required: a failure register written to satisfy a template stops
+    being a forecast, and boilerplate in a failure register is worse than
+    its absence. Include it when the plan is heavy or irreversible enough
+    that a named tripwire changes behaviour; omit it entirely otherwise —
+    never ship an empty or perfunctory one. Distinct from rule 24's Risk
+    Register: the register ranks *known* risks with mitigations; the
+    pre-mortem imagines the *whole plan dead* and reconstructs why.
+
 ---
 
 ## Quality Gates (remote CI by default)
@@ -357,6 +371,15 @@ complexity: lightweight
 
 {Optional: edge cases, decisions, links to related docs.}
 
+<!-- ## Pre-mortem — INCLUDE ONLY when the plan is heavy or irreversible
+     enough that a failure forecast changes behaviour (rule 25). OMIT
+     entirely otherwise — a perfunctory register is worse than none.
+     Shape: the four-part failure register from the premortem skill.
+1. Causes of death, ranked: {three, one paragraph each — mechanisms, not topics}
+2. Untested hidden dependency: {the assumption the plan never tests}
+3. Survivable-failure modification: {one concrete plan change}
+4. Tripwire: {metric + threshold + horizon that says cause #1 is materializing} -->
+
 <!-- ## Provenance — INCLUDE ONLY for source-derived / harvest / capability-adoption
      roadmaps (rule 19). OMIT entirely for internally-originated roadmaps —
      do NOT ship an empty Provenance section.

package/dist/cli-delegate/{chunk-22GCIJ2O.js → chunk-66OVWGB5.js}

@@ -2,7 +2,7 @@ import { createRequire as __acCreateRequire } from 'node:module'; const require
 import {
   DEFAULT_PROJECT_FILE,
   logger
-} from "./chunk-GB2FSUEE.js";
+} from "./chunk-LCFJIQ4J.js";
 
 // src/scripts/config/profiles.ts
 import * as fs2 from "node:fs";

package/dist/cli-delegate/{chunk-GB2FSUEE.js → chunk-LCFJIQ4J.js}

@@ -406,8 +406,40 @@ function load_agent_settings(options = {}) {
       _deep_merge(merged, layer);
     }
   }
+  _warn_removed_always_on_keys(merged);
   return merged;
 }
+var REMOVED_ALWAYS_ON_KEYS = [
+  "subagents.enabled",
+  "subagents.auto",
+  "subagents.host_capabilities",
+  "subagents.budget_routing",
+  "ai_team.enabled"
+];
+var _warnedRemovedAlwaysOnKeys = /* @__PURE__ */ new Set();
+function _readDottedSettingsPath(root, dotted) {
+  let node = root;
+  for (const part of dotted.split(".")) {
+    if (typeof node !== "object" || node === null || Array.isArray(node)) {
+      return void 0;
+    }
+    node = node[part];
+  }
+  return node;
+}
+function _warn_removed_always_on_keys(merged) {
+  for (const key of REMOVED_ALWAYS_ON_KEYS) {
+    if (_warnedRemovedAlwaysOnKeys.has(key)) {
+      continue;
+    }
+    if (_readDottedSettingsPath(merged, key) === void 0) {
+      continue;
+    }
+    _warnedRemovedAlwaysOnKeys.add(key);
+    process.stderr.write(`${key} was removed (always-on orchestration); ignored.
+`);
+  }
+}
 function* iter_setting_overrides(options = {}) {
   const project_path = options.project_path ?? null;
   const user_global_path = options.user_global_path ?? null;

package/dist/cli-delegate/{chunk-VSYT7OHO.js → chunk-LVRYBKTI.js}

@@ -343,6 +343,7 @@ function _build_config(raw, source_path) {
   const cli_call_budget = _build_cli_call_budget(
     _asDict(_getOr(raw, "cli_call_budget", {}))
   );
+  const quorum = _build_quorum(_get(raw, "quorum", "majority"));
   const necessity_classifier = _build_necessity_classifier(
     _asDict(_getOr(raw, "necessity_classifier", {}))
   );
@@ -368,6 +369,7 @@ function _build_config(raw, source_path) {
     "debate_gates"
   );
   const restate = _build_restate(_asDict(_getOr(raw, "restate", {})), "restate");
+  const critic_protocol = _build_critic_protocol(_getOr(raw, "critic_protocol", "legacy"));
   const decision_resolution = _build_decision_resolution(
     _asDict(_getOr(raw, "decision_resolution", {}))
   );
@@ -389,6 +391,7 @@ function _build_config(raw, source_path) {
     advisors,
     consensus_scoring: consensus,
     cli_call_budget,
+    quorum,
     necessity_classifier,
     model_downgrade,
     debate,
@@ -397,6 +400,7 @@ function _build_config(raw, source_path) {
     chairman,
     debate_gates,
     restate,
+    critic_protocol,
     decision_resolution,
     routing,
     low_impact,
@@ -598,6 +602,16 @@ function _build_restate(d, scope) {
   }
   return { enabled: Boolean(enabled) };
 }
+var _VALID_CRITIC_PROTOCOLS = /* @__PURE__ */ new Set(["legacy", "load_bearing"]);
+function _build_critic_protocol(v) {
+  if (v === void 0 || v === null) return "legacy";
+  if (!(_isStr(v) && _VALID_CRITIC_PROTOCOLS.has(v))) {
+    throw new CouncilConfigError(
+      `\`critic_protocol\`=${_pyRepr(v)} not in ${_sortedListRepr(_VALID_CRITIC_PROTOCOLS)}.`
+    );
+  }
+  return v;
+}
 var _VALID_RESOLUTION_MODES = /* @__PURE__ */ new Set([
   "agent",
   "council",
@@ -873,7 +887,7 @@ function _build_defaults(d) {
   if (!_isDict(d)) {
     throw new CouncilConfigError("`defaults` must be a mapping.");
   }
-  const mode = _get(d, "mode", "api");
+  const mode = _get(d, "mode", "auto");
   if (!(_isStr(mode) && _VALID_MODES.has(mode))) {
     throw new CouncilConfigError(
       `defaults.mode=${_pyRepr(mode)} not in ${_sortedListRepr(_VALID_MODES)}.`
@@ -1035,7 +1049,7 @@ function _build_cost_budget(d) {
   }
   return cb;
 }
-function _build_member(name, cfg, default_mode = "api") {
+function _build_member(name, cfg, default_mode = "auto") {
   if (!_VALID_PROVIDERS.has(name)) {
     throw new CouncilConfigError(
       `members.${name}: unknown provider; valid: ${_sortedListRepr(_VALID_PROVIDERS)}.`
@@ -1152,6 +1166,7 @@ function _build_member(name, cfg, default_mode = "api") {
     prompt_cache_ttl
   };
 }
+var _DEFAULT_CLI_CALLS_PER_DAY = 50;
 function _build_cli_call_budget(d) {
   if (!_isDict(d)) {
     throw new CouncilConfigError("`cli_call_budget` must be a mapping.");
@@ -1163,6 +1178,9 @@ function _build_cli_call_budget(d) {
     );
   }
   const caps = /* @__PURE__ */ new Map();
+  for (const provider of _VALID_PROVIDERS) {
+    caps.set(provider, _DEFAULT_CLI_CALLS_PER_DAY);
+  }
   for (const [provider, value] of Object.entries(raw_caps)) {
     if (!_VALID_PROVIDERS.has(provider)) {
       throw new CouncilConfigError(
@@ -1190,6 +1208,17 @@ function _build_cli_call_budget(d) {
   }
   return { max_calls_per_day: caps, warn_at };
 }
+function _build_quorum(raw) {
+  if (raw === "majority") {
+    return "majority";
+  }
+  if (_isInt(raw) && !_isBool(raw) && raw >= 1) {
+    return raw;
+  }
+  throw new CouncilConfigError(
+    `\`quorum\`=${_pyRepr(raw)} must be 'majority' or an integer >= 1 (got ${_pyTypeName(raw)}).`
+  );
+}
 function _build_advisor(name, cfg) {
   if (!_isDict(cfg)) {
     throw new CouncilConfigError(`advisors.${name}: must be a mapping.`);

package/dist/cli-delegate/{chunk-DVRS2KVR.js → chunk-MMXA5GY7.js}

@@ -4355,18 +4355,9 @@ var settingsSchema = external_exports.object({
     }).default({})
   }).default({}),
   subagents: external_exports.object({
-    enabled: external_exports.boolean().default(true).describe(
-      "Global master switch for the subagent layer. true (default) = available; false = fully disabled (the canonical kill-switch \u2014 no auto-dispatch, no routing, everything runs in-session)."
-    ),
-    auto: external_exports.enum(["off", "ask", "on"]).default("on").describe(
-      "Automatic-dispatch mode. off = subagents only via explicit command. ask = classify the task and ask once before dispatching. on = auto-dispatch delegable tasks (surface the choice in one line). Shipped default is on on subagent-capable hosts (flipped from ask 2026-07-09, ADR-117), off elsewhere; no-op where the host has no subagent primitive."
-    ),
     downshift: external_exports.boolean().default(true).describe(
       "Route delegable sub-tasks to the lowest-capable model tier (cost + speed via model downshift). false = every subagent runs on the session tier."
     ),
-    budget_routing: external_exports.enum(["ask", "auto", "off"]).default("ask").describe(
-      "Budget-aware cheap-request delegation (docs/contracts/budget-routing.md): pick the cheapest classifier-adequate tier WITH available budget (cost.budgets.per_tier); cheap exhausted but strong funded \u2192 strong tier; all exhausted \u2192 session model + notice. Work is never blocked to save money; the session model is never switched (delegation with a model override). ask = confirm the first budget-motivated downshift per session (default). auto = apply silently with telemetry. off = today's behavior (single-flip rollback). Distinct from downshift: that is the quality/cost knob, this is the resource-availability gate."
-    ),
     quota_arbitrage: external_exports.boolean().default(true).describe(
       "Prefer a separate quota-pool model for delegable sub-tasks where the host manifest reports one. Optional bonus only \u2014 identical behaviour (minus the quota win) where unsupported. Never load-bearing."
     ),
@@ -4377,9 +4368,6 @@ var settingsSchema = external_exports.object({
     }).default({}).describe(
       "Per-tier model map for downshift routing. Each empty value uses the tier runtime default (no vendor model baked in)."
     ),
-    host_capabilities: external_exports.object({}).passthrough().default({}).describe(
-      "Optional override of the host-capability manifest (subagent_spawn / parallel_spawn / status_polling / separate_quota_pool). Any field set wins; omitted fields fall back to the all-false safe default. Empty = the agent resolves the manifest from host knowledge."
-    ),
     implementer_model: external_exports.string().default("").describe(
       "Override the model the orchestrator dispatches to subagents that write code (e.g. claude-sonnet-4, gpt-5). Empty (default) = inherit the session's primary model \u2014 cheapest and usually right."
     ),
@@ -4399,14 +4387,11 @@ var settingsSchema = external_exports.object({
     )
   }),
   ai_team: external_exports.object({
-    enabled: external_exports.boolean().default(false).describe(
-      "Master switch for the /team cross-model review family (docs/contracts/ai-team-config.md). false (default) = commands are never suggested and every invocation refuses with an enable pointer. true = the family is live; individual gates (allow_delegate) still apply."
-    ),
     model: external_exports.string().default("auto").describe(
       "Model handed to the codex CLI. 'auto' (default) = pass no --model flag so the CLI's own default applies \u2014 tracks the subscription's current strongest model instead of pinning a stale ID. Any other value passes through verbatim as `--model <value>`."
     ),
     allow_delegate: external_exports.boolean().default(false).describe(
-      "Second opt-in for the only wrapper that delegates write access (/team:delegate). false (default) = delegate refuses even when ai_team.enabled is true. Both keys must be true before delegation is reachable."
+      "Second opt-in for the only wrapper that delegates write access (/team:delegate). false (default) = delegate refuses even when /team itself is available. Availability (codex CLI installed + authenticated) is necessary but not sufficient \u2014 this key must ALSO be true before delegation is reachable."
     ),
     max_calls_per_day: external_exports.number().int().min(0).default(50).describe(
       "Per-day cap on team calls, read against the EXISTING cli_call_budget openai bucket (~/.event4u/agent-config/cli-calls.json, daily UTC reset) \u2014 one subscription, one counter, never a parallel count. 0 = block all team calls."
@@ -4422,7 +4407,15 @@ var settingsSchema = external_exports.object({
         "Circuit-breaker bound: after this many CONSECUTIVE BLOCK verdicts in one session, a visible notice is injected exactly once and the managed layer stops re-blocking \u2014 the user decides, never an infinite Claude\u2194Codex loop. An ALLOW verdict resets the counter. Positive integer."
       )
     }).default({ managed: false, max_consecutive_blocks: 3 })
-  }).default({ enabled: false, model: "auto", allow_delegate: false, max_calls_per_day: 50, suppress_setup_hint: false, review_gate: { managed: false, max_consecutive_blocks: 3 } }),
+  }).default({ model: "auto", allow_delegate: false, max_calls_per_day: 50, suppress_setup_hint: false, review_gate: { managed: false, max_consecutive_blocks: 3 } }),
+  emergency: external_exports.object({
+    orchestration_halt: external_exports.boolean().default(false).describe(
+      "The one audited incident switch over the always-on orchestration stack (subagents, council, team). NOT an activation gate: false (default) = the stack runs normally. true = halted; arming requires no ceremony. Disarming (returning to false) requires orchestration_halt_justification to be a non-empty string. Both transitions emit one telemetry line."
+    ),
+    orchestration_halt_justification: external_exports.string().default("").describe(
+      "Required non-empty before orchestration_halt may return to false. Ignored while arming the halt."
+    )
+  }).default({ orchestration_halt: false, orchestration_halt_justification: "" }),
   onboarding: external_exports.object({
     onboarded: external_exports.boolean().default(false).describe(
       "Set to true once the developer has completed `agent-config setup`. The onboarding-gate rule blocks the first turn of every chat until this is true. Toggle back to false to re-trigger the wizard."

package/dist/cli-delegate/{chunk-JXARCGVN.js → chunk-MP3JT3IU.js}

@@ -1,7 +1,7 @@
 import { createRequire as __acCreateRequire } from 'node:module'; const require = globalThis.require ?? __acCreateRequire(import.meta.url);
 import {
   resolve_project_root
-} from "./chunk-GB2FSUEE.js";
+} from "./chunk-LCFJIQ4J.js";
 import {
   resolvePackageRoot
 } from "./chunk-MCJYCQPF.js";

package/dist/cli-delegate/{chunk-PQSAQKHW.js → chunk-NZ5N2XQL.js}

@@ -2,10 +2,10 @@ import { createRequire as __acCreateRequire } from 'node:module'; const require
 import {
   ZodFirstPartyTypeKind,
   settingsSchema
-} from "./chunk-DVRS2KVR.js";
+} from "./chunk-MMXA5GY7.js";
 import {
   main
-} from "./chunk-JXARCGVN.js";
+} from "./chunk-MP3JT3IU.js";
 import {
   RULE_SOURCE_REL
 } from "./chunk-BKO2AZ7Z.js";
@@ -48,7 +48,7 @@ import {
 import {
   find_project_root_with_anchor,
   load_agent_settings
-} from "./chunk-GB2FSUEE.js";
+} from "./chunk-LCFJIQ4J.js";
 import {
   event4u_root,
   legacy_xdg_root,