@event4u/agent-config 14.8.0 → 14.9.0

package/src/scripts/code_graph/cli.ts

@@ -77,8 +77,47 @@ function flag(argv: string[], name: string): string | null {
     return i >= 0 && i + 1 < argv.length ? (argv[i + 1] as string) : null;
 }
 
+/**
+ * Resolve the repository to operate on. `--root` > `AGENT_CONFIG_PROJECT_ROOT` >
+ * this module's own tree.
+ *
+ * The middle rung is the whole point, and its absence was a confused deputy:
+ * `_dispatch.bash` parses `--root` GLOBALLY, strips it from argv, and exports
+ * `AGENT_CONFIG_PROJECT_ROOT` (`:1655`) — then this engine read neither and fell
+ * back to its own module path. So `agent-config code-graph build --root <other
+ * repo>` silently indexed THIS package. The dispatcher made the decision and the
+ * engine substituted its own, which is why every consumer-facing invocation was
+ * wrong about which repository it had just read.
+ *
+ * `realpathSync` rather than `resolve` alone: a symlinked root must normalise to
+ * what will actually be walked, or the path a caller sees in the output is not
+ * the path that was indexed. Fail-loud on a root that does not exist — a silent
+ * fallback to the package tree is the defect this function exists to remove, and
+ * reintroducing it for a typo'd path would be the same bug wearing a smaller
+ * hat.
+ *
+ * Deliberately NOT touching `DEFAULT_CACHE`, which is anchored to `REPO_ROOT`
+ * for the same reason and is a separate decision: it is a WRITE path, with its
+ * own ownership, cleanup, concurrency and multi-repo namespacing questions.
+ * Council 2026-08-22, both seats, explicitly out of scope here.
+ */
+function resolveRoot(argv: string[]): string {
+    const explicit = flag(argv, '--root');
+    const fromDispatcher = process.env['AGENT_CONFIG_PROJECT_ROOT'];
+    const candidate =
+        explicit ?? (fromDispatcher !== undefined && fromDispatcher !== '' ? fromDispatcher : REPO_ROOT);
+    try {
+        return fs.realpathSync(path.resolve(candidate));
+    } catch {
+        // Returned unresolved on purpose: every caller checks existence and
+        // prints the path it was given, which is more useful in the error than
+        // a normalised form of a path that does not exist.
+        return path.resolve(candidate);
+    }
+}
+
 async function cmdBuild(argv: string[]): Promise<number> {
-    const root = flag(argv, '--root') ?? REPO_ROOT;
+    const root = resolveRoot(argv);
     const out = argv.includes('--out') ? flag(argv, '--out') : DEFAULT_CACHE;
     if (!fs.existsSync(root)) {
         process.stderr.write(`code-graph: root not found: ${root}\n`);
@@ -122,7 +161,7 @@ function cmdValidate(argv: string[]): number {
 
 function resolveGraph(argv: string[]): { g: LoadedGraph; note: string } | { err: string } {
     const explicit = flag(argv, '--graph');
-    const root = flag(argv, '--root') ?? REPO_ROOT;
+    const root = resolveRoot(argv);
     if (explicit) {
         if (!fs.existsSync(explicit)) return { err: `graph not found: ${explicit}` };
         return { g: loadGraph(explicit, path.relative(REPO_ROOT, explicit)), note: '' };
@@ -179,7 +218,7 @@ function cmdQuery(kind: 'query' | 'explain' | 'affected' | 'path', argv: string[
     const positional = argv.filter((a) => !a.startsWith('--') && !isFlagValue(argv, a));
     const budget = budgetOf(argv);
     if (kind === 'affected' && flag(argv, '--since')) {
-        const root = flag(argv, '--root') ?? REPO_ROOT;
+        const root = resolveRoot(argv);
         const seeds = changedNodeSeeds(r.g, root, flag(argv, '--since') as string);
         const merged: QueryResult = { source: r.g.source, seeds: [], lines: [], truncated: false, recommended_reads: [] };
         const perSeed: QueryResult[] = [];
@@ -216,7 +255,7 @@ function isFlagValue(argv: string[], token: string): boolean {
 }
 
 function cmdDetect(argv: string[]): number {
-    const root = flag(argv, '--root') ?? REPO_ROOT;
+    const root = resolveRoot(argv);
     if (flag(argv, '--format') === 'json') {
         // Machine-readable three-state verdict (Phase 2). Existing human-text
         // output below is unchanged for existing consumers.
@@ -316,7 +355,7 @@ function runBudgetedBuild(root: string, out: string, update: boolean, budgetSeco
 }
 
 function cmdRefresh(argv: string[]): number {
-    const root = flag(argv, '--root') ?? REPO_ROOT;
+    const root = resolveRoot(argv);
     const out = flag(argv, '--out') ?? DEFAULT_CACHE;
     const budgetSeconds = budgetSecondsOf(argv);
     if (!fs.existsSync(root)) {

package/src/scripts/code_graph/extract.ts

@@ -306,6 +306,93 @@ function extractTsJs(root: TsNode, file: string, out: FileExtract): void {
                 if (body) for (const c of namedChildren(body)) walk(c, classId, id);
                 return;
             }
+            // Modern TS declares most of its functions as bindings, not as
+            // `function` statements. Before this case the extractor produced
+            // NO node for `export const f = () => {}` or
+            // `const g = function () {}`, which is the measured cause of the
+            // 170-vs-13,428 TS/PHP symbol gap: the six handled kinds simply do
+            // not cover the dominant declaration form.
+            //
+            // Scoped deliberately to bindings whose value IS a function. A
+            // `const x = 3` is data, not a symbol this graph answers questions
+            // about, and emitting a node per constant would inflate the count
+            // without improving recall — which is the cosmetic-improvement
+            // failure the phase falsifier exists to catch.
+            case 'lexical_declaration':
+            case 'variable_declaration': {
+                for (const d of namedChildren(n)) {
+                    if (d.type !== 'variable_declarator') continue;
+                    const value = d.childForFieldName('value');
+                    if (!value || (value.type !== 'arrow_function' && value.type !== 'function_expression')) {
+                        continue;
+                    }
+                    const nm = nameField(d) ?? '<anon>';
+                    const id = classId ? `${classId}::${nm}` : `${file}#${nm}`;
+                    out.nodes.push({
+                        id,
+                        label: nm,
+                        kind: classId ? 'method' : 'function',
+                        source_file: file,
+                        source_location: loc(d),
+                    });
+                    out.rawEdges.push({
+                        sourceId: classId ?? fileId,
+                        relation: 'member',
+                        targetName: nm,
+                        confidenceHint: 'EXTRACTED',
+                    });
+                    // Walk the body under the BINDING's scope, so a call inside
+                    // an arrow function attributes to that function rather than
+                    // to the enclosing file — the same treatment
+                    // `function_declaration` gets.
+                    //
+                    // `walk(body)` and not `namedChildren(body)`: an
+                    // expression-bodied arrow (`() => helper()`) has the CALL
+                    // itself as its body, so descending one level first steps
+                    // straight past it and the edge is lost. The sibling cases
+                    // can iterate children because a `statement_block` is never
+                    // the interesting node; here it often is.
+                    const body = value.childForFieldName('body');
+                    if (body) walk(body, classId, id);
+                }
+                // Non-function declarators still need walking for their calls.
+                for (const d of namedChildren(n)) {
+                    if (d.type !== 'variable_declarator') continue;
+                    const value = d.childForFieldName('value');
+                    if (value && (value.type === 'arrow_function' || value.type === 'function_expression')) {
+                        continue;
+                    }
+                    walk(d, classId, scopeId);
+                }
+                return;
+            }
+            // `class C { m = () => 1 }` — a class property holding a function.
+            // Emitted as a `method`, the same kind `method_definition` emits,
+            // because it is one to every caller.
+            case 'public_field_definition': {
+                const value = n.childForFieldName('value');
+                const nm = nameField(n) ?? '<anon>';
+                if (value && (value.type === 'arrow_function' || value.type === 'function_expression')) {
+                    const id = classId ? `${classId}::${nm}` : `${file}#${nm}`;
+                    out.nodes.push({
+                        id,
+                        label: nm,
+                        kind: 'method',
+                        source_file: file,
+                        source_location: loc(n),
+                    });
+                    out.rawEdges.push({
+                        sourceId: classId ?? fileId,
+                        relation: 'member',
+                        targetName: nm,
+                        confidenceHint: 'EXTRACTED',
+                    });
+                    const body = value.childForFieldName('body');
+                    if (body) walk(body, classId, id);
+                    return;
+                }
+                break;
+            }
             case 'new_expression': {
                 const ctor = n.childForFieldName('constructor');
                 if (ctor && (ctor.type === 'identifier' || ctor.type === 'type_identifier' || ctor.type === 'member_expression'))

package/src/scripts/council_cli.ts

@@ -106,7 +106,8 @@ import {
     renderPostureLines,
     type FallbackPosture,
 } from './_lib/council_fallback_posture.js';
-import { buildFallbackOptions } from './_lib/council_fallback_wiring.js';
+import { wireCouncilFallback } from './_lib/council_fallback_wiring.js';
+import { BILLING_SUBCOMMANDS, billingArgSpec, handleBillingCommand, printBillingGate } from './_lib/billing_grant_cli.js';
 import {
     type ClassificationResult,
     type SizeFitVerdict,
@@ -880,19 +881,11 @@ function build_members(settings: Dict, opts: BuildMembersOptions = {}): External
                 'prints the resolved path.',
         );
     }
-    if (fallback_out !== null) {
-        // Absent/malformed block means `false`, never a throw. Why quota
-        // fall-through is opt-in: `FallbackPolicy.apiOnQuota`.
-        const fallback_cfg = _isDict(ai['fallback']) ? (ai['fallback'] as Dict) : {};
-        fallback_out.options = buildFallbackOptions({
-            apiOnQuota: _pyBool(fallback_cfg['api_on_quota']),
-            hasApiRung: (p) => _API_PROVIDERS.has(p),
-            memberConfig: (p) => ((members_cfg[p] as Dict) || {}) as Dict,
-            modelOverride: (p) => (overrides[p] as string | undefined) ?? null,
-            constructApi: _construct_api_member,
-            emit: appendEvent,
-        });
-    }
+    wireCouncilFallback(fallback_out, ai, {
+        repoRoot: REPO_ROOT, isDict: _isDict, membersCfg: members_cfg, overrides,
+        hasApiRung: (p) => _API_PROVIDERS.has(p),
+        constructApi: _construct_api_member, emit: appendEvent,
+    });
     return members;
 }
 
@@ -2668,6 +2661,7 @@ function cmd_run(
     quorum_out.result = post_run.quorum;
     skipped.push(...post_run.absent);
     recordRoundObservations(REPO_ROOT, members, responses, classifyCliFailure);
+    printBillingGate(fallback_out.options?.parked ?? [], { repoRoot: REPO_ROOT, stdout: _stdout, stderr: _stderr });
     // Attendance telemetry for the reading that actually decided the pass.
     // `post_run.absent` is the mid-flight set only; the construction-time
     // skips already went out under `phase: 'pre_run'` from `build_members`,
@@ -3509,7 +3503,7 @@ interface Args {
     json: boolean;
 }
 
-const _SUBCOMMANDS = ['estimate', 'run', 'debate', 'render', 'replay', 'quota', 'shadow-report', 'status'];
+const _SUBCOMMANDS = ['estimate', 'run', 'debate', 'render', 'replay', 'quota', 'shadow-report', 'status', ...BILLING_SUBCOMMANDS];
 
 const _TOP_USAGE =
     `usage: ${_PROG} [-h]\n` +
@@ -3698,7 +3692,7 @@ function _specsFor(cmd: string): { positionals: string[]; opts: OptSpec[]; requi
                 requiredOpts: [],
             };
         default:
-            return { positionals: [], opts: [], requiredOpts: [] };
+            return billingArgSpec(cmd) ?? { positionals: [], opts: [], requiredOpts: [] };
     }
 }
 
@@ -3902,6 +3896,8 @@ function main(argv: string[] | null = null): number {
         if (args.cmd === 'status') {
             return cmd_status(args);
         }
+        const billing = handleBillingCommand(String(args.cmd), _getattr(args, 'run_id', null), { repoRoot: REPO_ROOT, stdout: _stdout, stderr: _stderr });
+        if (billing !== null) return billing;
     } catch (exc) {
         if (exc instanceof CouncilDisabledError) {
             _stderr(`❌  council:${args.cmd}: ${exc.message}\n`);

package/src/scripts/detect_ai_tells.ts

@@ -111,6 +111,93 @@ export function scanHiddenUnicode(text: string): HiddenUnicodeFinding[] {
   }));
 }
 
+/** One codepoint the strip considered, and what it decided. */
+export interface CarrierRecord {
+  /** `U+XXXX` form, so a record is readable without decoding. */
+  codepoint: string;
+  /** The class `_classify` returned — never a second class list. */
+  cls: string;
+  /** Codepoint index (not UTF-16 index) in the input. */
+  offset: number;
+  disposition: "removed" | "preserved";
+  /** Why it was preserved. `null` when removed. */
+  reason: string | null;
+}
+
+export interface CarrierStripResult {
+  out: string;
+  removed: number;
+  preserved: number;
+  records: CarrierRecord[];
+}
+
+/**
+ * Remove hidden-Unicode carriers from OUTPUT prose — context-aware, opt-in.
+ *
+ * ## Why this is not `_sanitize`
+ *
+ * `lint_hidden_unicode._sanitize` drops every `_classify`-flagged codepoint
+ * unconditionally and then applies `NFKC`. That is correct for its own callers —
+ * it repairs a source FILE — and it is wrong for a deliverable, because
+ * `_ZERO_WIDTH` contains `0x200C` and `0x200D`. A blind pass therefore destroys
+ * emoji ZWJ sequences and complex-script joiners, which are legitimate content.
+ * This function keeps `_classify` as the single source of truth for *what is a
+ * candidate* and adds exactly one predicate for *whether to remove it*.
+ *
+ * ## The predicate
+ *
+ * A flagged codepoint is removed only when the codepoints on BOTH sides are
+ * ASCII (`< 0x80`) or absent. A carrier between two ASCII characters is doing
+ * no linguistic work; one adjacent to any non-ASCII character might be a
+ * joiner, so it stays.
+ *
+ * Deliberately conservative, and the asymmetry is the point: the failure mode
+ * is a preserved carrier, never a corrupted word. That makes this a HYGIENE
+ * pass and not a security control — the injection vector is covered on the
+ * INGESTION side by `scanHiddenUnicode`, whose contract is to warn and never
+ * strip, and which this function does not touch.
+ *
+ * Not a detector-evasion tool: it removes invisible characters and alters no
+ * visible prose. That exclusion is permanent.
+ */
+export function stripCarrierUnicode(text: string): CarrierStripResult {
+  const cps = [...text].map((ch) => ch.codePointAt(0) ?? 0);
+  const kept: string[] = [];
+  const records: CarrierRecord[] = [];
+  let removed = 0;
+  let preserved = 0;
+
+  for (let i = 0; i < cps.length; i += 1) {
+    const cp = cps[i] as number;
+    const cls = _classify(cp);
+    if (cls === null) {
+      kept.push(String.fromCodePoint(cp));
+      continue;
+    }
+    // `undefined` at either end is "absent", which counts as ASCII-side: a
+    // carrier at a boundary has no neighbour to be joining.
+    const before = cps[i - 1];
+    const after = cps[i + 1];
+    const asciiSide = (n: number | undefined): boolean => n === undefined || n < 0x80;
+    const strip = asciiSide(before) && asciiSide(after);
+    records.push({
+      codepoint: `U+${cp.toString(16).toUpperCase().padStart(4, "0")}`,
+      cls,
+      offset: i,
+      disposition: strip ? "removed" : "preserved",
+      reason: strip ? null : "adjacent to a non-ASCII codepoint — may be a joiner",
+    });
+    if (strip) {
+      removed += 1;
+    } else {
+      preserved += 1;
+      kept.push(String.fromCodePoint(cp));
+    }
+  }
+
+  return { out: kept.join(""), removed, preserved, records };
+}
+
 interface Thresholds {
   maxHard: number;
   maxScore: number;

package/src/scripts/dispatch_r2_reviewer.ts

@@ -698,6 +698,44 @@ function isoSeconds(d: Date): string {
     return d.toISOString().replace(/\.\d{3}Z$/, 'Z');
 }
 
+/**
+ * The return-envelope POINTER — one line-shaped field, delivered on the
+ * dominant dispatch path.
+ *
+ * ## Why a pointer and not the contract
+ *
+ * 240 chars is not a round number: it is `MAX_RESPONSE_LINE_CHARS`
+ * (`_lib/subagent_response.ts`), so the pointer is one line-shaped field by the
+ * contract's own measure. Inlining the shape would put a fourth copy of it in
+ * the tree, and a fourth copy is how the contract came to have three
+ * inconsistent states.
+ *
+ * ## Why THIS dispatcher
+ *
+ * `road-to-subagent-envelope-adoption` step 1.1 says the dominant path, "and it
+ * does not fan out to both". The dominant path is **not answerable from the
+ * ledger** — 1,725 of 1,845 post-split stops carry a null `agent_type`, which is
+ * the 8 % start↔stop join rate recorded elsewhere, so no stop can be attributed
+ * to a dispatcher. Decided on invocation surface instead: this file is wired
+ * into `/create-pr` and read by three gates (`check_completion_review`,
+ * `lint_evidence_artifacts`, `probe_review_binding_drift`), while
+ * `team_dispatch` sits behind an ai_team host flag and one `/team delegate`
+ * command. It is also the leg with ZERO prior contract hits, so the dominant
+ * path and the absent one are the same file.
+ *
+ * ## What it does not change
+ *
+ * The findings table stays the artefact — the gates read the file, not the
+ * envelope. Per the four-boundary separation in the response contract, this
+ * touches the DELIVERY boundary only: what the final message is. It adds no
+ * lifecycle status and does not alter the body schema.
+ */
+export const RETURN_ENVELOPE_POINTER =
+    'Final message = the return envelope and nothing else: ' +
+    '{summary, handoff, confidence, findings, risks}. Shape + the ' +
+    'write-to-disk-first rule: contexts/execution/subagent-response-contract.md. ' +
+    'The findings table stays a file.';
+
 function reviewerPrompt(args: {
     slug: string;
     headSha: string;
@@ -781,6 +819,10 @@ function reviewerPrompt(args: {
         `**Honest-null:** 0 findings, scope ${args.scopeHash}, reviewed <YYYY-MM-DD>`,
         '```',
         '',
+        '## Return channel',
+        '',
+        RETURN_ENVELOPE_POINTER,
+        '',
     ].join('\n');
 }
 

package/src/scripts/generate_capabilities_index.ts

@@ -268,6 +268,45 @@ function _getStr(d: Dict, key: string, dflt: string): string {
     return String(v);
 }
 
+/**
+ * What the runtime hooks actually DO, in four states and no others.
+ *
+ * The four: `detects` (it looks and reports nothing a host reads), `warns` (it
+ * reports through the host, and the host may ignore it), `blocks` (it can
+ * refuse), and `not-enforceable-on-host` (the slot does not exist there).
+ *
+ * WHY THIS BLOCK EXISTS. `CAPABILITIES.yaml` carried no prompt-injection entry
+ * at all, and saying nothing is a different defect from saying too much: a
+ * reader looking for injection coverage found absence, which reads as "not
+ * covered" rather than as "a warn-only scanner exists and ships off". This is
+ * the constructive half of a source proposal whose other half — "remove claims
+ * that equate a warn-hook with prevention" — had nothing to remove.
+ *
+ * `blocks` is never legal for the injection scanner on ANY host, and the value
+ * is DERIVED from `hook_manifest.yaml` rather than typed here: `fail_closed:
+ * false` plus `severity: advisory` is what makes it `warns`, so a manifest edit
+ * that gave it teeth would have to change this row too. A hand-written `blocks`
+ * would be the coverage inflation `untrusted-input-defense` refuses, one file
+ * over.
+ */
+function _runtime_enforcement_lines(): string[] {
+    return [
+        'runtime_enforcement:',
+        '  # Four states, and no others: detects | warns | blocks | not-enforceable-on-host.',
+        '  # Derived from src/scripts/hook_manifest.yaml (fail_closed + severity), never',
+        '  # hand-written — a manifest row that gained teeth would have to change this one.',
+        '  - hook: injection-scan',
+        '    slot: post_tool_use',
+        '    state: warns',
+        '    default: off',
+        '    blocks_legal_on_any_host: false',
+        '    reason: ' +
+            _scalar(
+                'fail_closed: false + severity: advisory in hook_manifest.yaml — it reports through the host and cannot refuse, so it warns and never blocks. Ships enabled: false; the detector was measured at 99.00% recall and a 0.85% false-positive rate over internal/bench/corpora/encoding-channels/ before any default flip is even discussable.<!-- claim:injection-scan-corpus-rates -->',
+            ),
+    ];
+}
+
 export function build(): string {
     const packs = _load_packs();
     const skillMap = _skill_packs();
@@ -360,6 +399,8 @@ export function build(): string {
         }
     }
     lines.push('');
+    lines.push(..._runtime_enforcement_lines());
+    lines.push('');
     lines.push('gaps:');
     if (gaps.length > 0) {
         const sortedGaps = [...gaps].sort((a, b) => (a.id < b.id ? -1 : a.id > b.id ? 1 : 0));

package/src/scripts/git_authorization_hook.ts

@@ -153,6 +153,19 @@ export type GitOp =
   | "branch"
   | "pr-create"
   | "pr-merge"
+  /**
+   * Enabling auto-merge. Its OWN op, not a plain merge and not nothing.
+   *
+   * Irreversible in the same sense a merge is: it commits the outcome to a
+   * condition the agent does not control, and the merge then happens with
+   * nobody in the loop. Before this split, `gh pr merge --auto` classified
+   * identically to a plain merge (the pattern is `\bgh\s+pr\s+merge\b`, and
+   * `--auto` is invisible to it), and the GraphQL mutation classified as
+   * NOTHING — `ghApiWrite()` needs a REST path and a write method, and
+   * `gh api graphql -f query='mutation{enablePullRequestAutoMerge...}'` has
+   * neither.
+   */
+  | "pr-merge-auto"
   | "tag"
   | "release"
   | "publish";
@@ -163,6 +176,7 @@ export const ALL_OPS: readonly GitOp[] = [
   "branch",
   "pr-create",
   "pr-merge",
+  "pr-merge-auto",
   "tag",
   "release",
   "publish",
@@ -185,8 +199,29 @@ const PHRASES: ReadonlyArray<{ op: GitOp; re: RegExp }> = [
     op: "pr-create",
     re: /\b(erstell(e|en)?|mach(e|en)?|leg(e|en)?\s+an|(er)?(ö|oe)ffne|open|create|raise|aufmachen)\b[^.\n]{0,30}\b(pr|pull[- ]request)\b|\b(pr|pull[- ]request)\b[^.\n]{0,20}\b(erstellen|aufmachen|anlegen|(er)?(ö|oe)ffnen)\b/i,
   },
-  // `merge` as an ACTION, never as the noun in "merge conflict" / "merge commit".
-  { op: "pr-merge", re: /\b(merge|merg(e|en|st|t)|zusammenf(ü|ue)hren|reinmergen)\b(?!\s*[- ]?(conflict|konflikt|commit|base|queue|state|status))/i },
+  // `merge` as an ACTION, never as the noun in "merge conflict" / "merge commit",
+  // and never under a NEGATION.
+  //
+  // The negation half closes a measured defect: probed directly against
+  // `classifyAuthorization`, "nicht mergen", "don't merge this" and
+  // "never auto-merge" each returned ["pr-merge"] — a prompt saying DO NOT
+  // MERGE authorized a merge. The noun-sense lookahead was never a negator.
+  //
+  // The lookBEHIND is line-scoped by `[^.!?\n]` on purpose, the same scoping
+  // `turn_end_gate_hook.ts` uses for its own negation exclusion (`:330`) and
+  // its negated-claim window (`:490-493`). A sentence boundary ends a
+  // negation's reach: "Do not push. Merge PR #12." is two instructions, and a
+  // guard that let the first suppress the second would silently stop
+  // authorizing merges the user DID order — the failure mode that is worse
+  // than the defect, because nothing happens and nothing says why.
+  //
+  // Reusing that vocabulary rather than inventing a third is deliberate: two
+  // negation vocabularies in one tree drift, and the drift is invisible until
+  // a prompt lands in the gap between them.
+  {
+    op: "pr-merge",
+    re: /(?<!\b(nicht|nichts|kein(e|en|em|er|es)?|niemals|nie|no|not|dont|don't|never|without|ohne)\b[^.!?\n]{0,30})\b(merge|merg(e|en|st|t)|zusammenf(ü|ue)hren|reinmergen)\b(?!\s*[- ]?(conflict|konflikt|commit|base|queue|state|status))/i,
+  },
   // A tag is an ACTION here — bare "Tag" is the German word for day, and
   // "Version" is an ordinary noun. Both authorized a BLOCK op before this.
   { op: "tag", re: /\b(tagge(n|st)?|tag\s+(setzen|anlegen|erstellen)|git\s+tag|--tags|--follow-tags)\b/i },
@@ -267,10 +302,20 @@ export function isAffirmative(prose: string): boolean {
 }
 
 /** Executable commands a user may paste, mapped to the op they authorize. */
-const PASTED_COMMANDS: ReadonlyArray<{ op: GitOp; re: RegExp }> = [
+const PASTED_COMMANDS: ReadonlyArray<{ op: GitOp | null; re: RegExp }> = [
   { op: "publish", re: /\bnpm\s+publish\b/i },
   { op: "tag", re: /\bgit\s+push\s+[^\n]*--tags\b|\bgit\s+tag\s+-a\b/i },
   { op: "release", re: /\bgh\s+release\s+create\b/i },
+  // ORDER IS LOAD-BEARING: the auto variants precede the plain merge, because
+  // the first match wins and `gh pr merge 12 --auto` also matches the plain
+  // pattern. Same rule the block-side table states for `git push --tags`.
+  //
+  // The DE-ESCALATING forms come first of all, and they map to no op at all:
+  // `--disable-auto` and `disablePullRequestAutoMerge` turn the capability OFF.
+  // Requiring merge authorization to switch auto-merge off is a live deadlock,
+  // and it is why this cannot ship after the split rather than with it.
+  { op: null, re: /\bgh\s+pr\s+merge\b[^\n]*--disable-auto\b|\bdisablePullRequestAutoMerge\b/i },
+  { op: "pr-merge-auto", re: /\bgh\s+pr\s+merge\b[^\n]*--auto\b|\benablePullRequestAutoMerge\b/i },
   { op: "pr-merge", re: /\bgh\s+pr\s+merge\b/i },
   { op: "pr-create", re: /\bgh\s+pr\s+create\b/i },
   { op: "push", re: /\bgit\s+push\b/i },
@@ -355,10 +400,15 @@ export function classifyAuthorization(prompt: string): {
     }
     for (const line of _commandLines(fence)) {
       for (const { op, re } of PASTED_COMMANDS) {
-        if (re.test(line)) {
+        if (!re.test(line)) continue;
+        // FIRST MATCH WINS, and the table is ordered most-specific-first. A
+        // de-escalating form carries `op: null` — it matched, so no later
+        // pattern may claim the line, and it authorizes nothing.
+        if (op !== null) {
           authorized.add(op);
           evidence[op] = `pasted command: "${line.slice(0, 80)}"`;
         }
+        break;
       }
     }
   }

package/src/scripts/hooks/block_unauthorized_git.ts

@@ -92,6 +92,9 @@ export const BLOCK_OPS: ReadonlySet<GitOp> = new Set<GitOp>([
   "tag",
   "release",
   "pr-merge",
+  // Enabling auto-merge commits the outcome to a condition the agent does not
+  // control, so it is blocked on the same terms as the merge it schedules.
+  "pr-merge-auto",
 ]);
 
 /** Recoverable operations — warned without a this-turn authorization. */
@@ -162,10 +165,25 @@ const COMMAND_OPS: ReadonlyArray<{ op: GitOp; re: RegExp }> = [
     op: "tag",
     re: new RegExp(`${P}git\\s+${G}push\\b[^\\n;|&]*(--tags|--follow-tags|refs\\/tags\\/)`, "i"),
   },
+  // ORDER IS LOAD-BEARING, per this table's own rule that the most specific
+  // pattern wins: `gh pr merge 12 --auto` matches the plain merge pattern too.
+  //
+  // The GraphQL mutation needs its OWN pattern rather than a widening of
+  // `ghApiWrite()`. That helper requires BOTH a REST-shaped path and an
+  // explicit write method, and `gh api graphql -f query='mutation{...}'` has
+  // neither — widening it to reach this would loosen every other op that uses
+  // it, which is a bigger change than the defect.
+  {
+    op: "pr-merge-auto",
+    re: new RegExp(
+      `${P}gh\\s+pr\\s+merge\\b(?=[^\\n]*--auto\\b)|${P}gh\\s+api\\b(?=[^\\n]*enablePullRequestAutoMerge\\b)`,
+      "i",
+    ),
+  },
   {
     op: "pr-merge",
     re: new RegExp(
-      `${P}gh\\s+pr\\s+merge\\b|${ghApiWrite("\\/pulls\\/\\d+\\/merge\\b").source}`,
+      `${P}gh\\s+pr\\s+merge\\b(?![^\\n]*--(auto|disable-auto)\\b)|${ghApiWrite("\\/pulls\\/\\d+\\/merge\\b").source}`,
       "i",
     ),
   },

package/src/scripts/hooks/run_continuation_hook.ts

@@ -129,6 +129,7 @@
  */
 
 import * as fs from 'node:fs';
+import { revokeBillingGrant } from '../_lib/billing_grant.js';
 import * as path from 'node:path';
 import { fileURLToPath, pathToFileURL } from 'node:url';
 
@@ -599,12 +600,13 @@ function writeState(file: string, state: RunState): boolean {
  * stamp with it. Adoption is a fact this read already established; it is passed in
  * rather than re-derived from a condition that cannot express it.
  */
-function clearRunState(
+export function clearRunState(
     workspaceRoot: string,
     runId: string,
     keyedFile: string,
     adoptedLegacy: boolean,
 ): void {
+    revokeBillingGrant(workspaceRoot, runId); // the run IS the grant's TTL
     try {
         fs.rmSync(keyedFile, { force: true });
     } catch {