@event4u/agent-config 9.7.0 → 9.9.0

package/src/scripts/context_hygiene_hook.ts

@@ -36,6 +36,7 @@ import { fileURLToPath, pathToFileURL } from "node:url";
 // `agents/runtime/state/.dispatcher.lock` discipline (hook-architecture-v1.md
 // § Concurrency, Phase 7.4).
 import { atomic_write_json } from "./hooks/state_io.js";
+import { readHookStdin } from "./hooks/hook_stdin.js";
 
 // NOTE: the Python docstring says `agents/runtime/state/`, but the code
 // constant is `agents/state/`. Replicated verbatim — this is a latent
@@ -252,11 +253,7 @@ function parse_args(argv: string[]): ParsedArgs {
 }
 
 function _readStdin(): string {
-  try {
-    return fs.readFileSync(0, "utf-8");
-  } catch {
-    return "";
-  }
+  return readHookStdin();
 }
 
 export function main(argv?: string[]): number {
@@ -264,7 +261,13 @@ export function main(argv?: string[]): number {
   return run(_readStdin(), { consumer_root: process.cwd(), verbose: args.verbose });
 }
 
+// Bundle-safety: never auto-run when inlined into an esbuild bundle, where
+// every module shares the bundle's `import.meta.url` (see cmd_migrate.ts).
+declare const __AGENT_CONFIG_BUNDLE__: boolean | undefined;
 function _isCliEntry(): boolean {
+    if (typeof __AGENT_CONFIG_BUNDLE__ !== 'undefined' && __AGENT_CONFIG_BUNDLE__) {
+        return false;
+    }
     if (process.argv[1] === undefined) {
         return false;
     }

package/src/scripts/council_cli.ts

@@ -117,6 +117,15 @@ import {
 } from './ai_council/consensus.js';
 import * as _shadow from './ai_council/shadow_dispatch.js';
 import * as _lowimpact from './ai_council/low_impact.js';
+import {
+    apply_chairman_override,
+    assign_stances,
+    build_blind_labels,
+    OUTSIDER_STANCE_NAME,
+    parse_chairman_override,
+    render_deanonymization_block,
+    with_chairman_fields,
+} from './ai_council/blind_review.js';
 
 // ── argparse-style exit plumbing ────────────────────────────────────
 // Mirror CPython argparse: a `prog: error: …` line on stderr + exit 2.
@@ -1168,16 +1177,36 @@ function _maybe_run_chairman(
             response: null,
         };
     }
-    // Transcript WITH identities (the chairman judges attributed positions) +
-    // the lens synthesis template as the authoring instruction.
-    const transcript = responses
-        .filter((r) => r.error === null && r.text.trim() !== '')
-        .map((r) => `## ${r.provider} - ${r.model}\n\n${r.text.trim()}`)
-        .join('\n\n---\n\n');
-    const synth_prompt =
-        `You are the council CHAIRMAN. You did not deliberate. Author the synthesis ` +
-        `of the attributed member positions below, following the template.\n\n` +
-        `${synthesis_template(question.mode)}\n\n---\n\n${transcript}`;
+    // road-to-council-blind-review Phase 1 (Ü1): --blind-chairman shuffles the
+    // transcript deterministically (seeded by the question text) and strips
+    // identity to Response-A/B/… labels, reusing the EXISTING
+    // `consensus.anonymize_responses` seam. The default path below (transcript
+    // WITH identities — "the chairman judges attributed positions") is
+    // byte-identical to today.
+    const blind = args.blind_chairman === true;
+    let transcript: string;
+    let blind_label_to_source: Map<string, string> | null = null;
+    const successful = responses.filter((r) => r.error === null && r.text.trim() !== '');
+    if (blind) {
+        const pairs: Array<[string, string]> = successful.map((r) => [`${r.provider}:${r.model}`, r.text]);
+        const built = build_blind_labels(question.user_prompt, pairs);
+        transcript = built.transcript;
+        blind_label_to_source = built.label_to_source;
+    } else {
+        transcript = successful
+            .map((r) => `## ${r.provider} - ${r.model}\n\n${r.text.trim()}`)
+            .join('\n\n---\n\n');
+    }
+    // Ü3: --chairman-fields appends the two mandatory trailing sections.
+    const template =
+        args.chairman_fields === true ? with_chairman_fields(synthesis_template(question.mode)) : synthesis_template(question.mode);
+    const synth_prompt = blind
+        ? `You are the council CHAIRMAN. You did not deliberate. Author the synthesis ` +
+          `of the anonymized member positions below (labels A–E). Do not guess ` +
+          `identities.\n\n${template}\n\n---\n\n${transcript}`
+        : `You are the council CHAIRMAN. You did not deliberate. Author the synthesis ` +
+          `of the attributed member positions below, following the template.\n\n` +
+          `${template}\n\n---\n\n${transcript}`;
     const synthQ = new CouncilQuestion({
         mode: question.mode,
         user_prompt: synth_prompt,
@@ -1197,7 +1226,13 @@ function _maybe_run_chairman(
             response: r,
         };
     }
-    return { member: sel.member, annotation: sel.annotation, text: r.text, response: r };
+    // Blind is only at decision time — de-anonymization lands in the audit
+    // artifact immediately after the verdict, never fed back into a prompt.
+    const final_text =
+        blind && blind_label_to_source !== null
+            ? `${r.text.trim()}\n\n${render_deanonymization_block('### De-anonymization (post-verdict)', blind_label_to_source)}`
+            : r.text;
+    return { member: sel.member, annotation: sel.annotation, text: final_text, response: r };
 }
 
 function _maybe_run_peer_review(
@@ -1840,6 +1875,15 @@ function _apply_solo_dispatch(members: ExternalAIClient[]): [ExternalAIClient[],
     ];
 }
 
+/** `--chairman` value → parsed override, or `ArgumentTypeError` on malformed input. */
+function _resolved_chairman_override(raw: string | null) {
+    try {
+        return parse_chairman_override(raw);
+    } catch (exc) {
+        throw new ArgumentTypeError(exc instanceof Error ? exc.message : String(exc));
+    }
+}
+
 function cmd_run(
     args: Args,
     opts: { settings?: Dict | null; members?: ExternalAIClient[] | null; table?: PriceTable | null } = {},
@@ -1850,7 +1894,12 @@ function cmd_run(
     if (settings === null) {
         settings = load_settings();
     }
-    const ai_cfg = _isDict(settings) ? ((settings['ai_council'] as Dict) || {}) : {};
+    // road-to-council-blind-review Phase 1: `--chairman host|auto|member:NAME`
+    // is a pure runtime override, never a config write.
+    const ai_cfg = apply_chairman_override(
+        _isDict(settings) ? ((settings['ai_council'] as Dict) || {}) : {},
+        _resolved_chairman_override(_getattr<string | null>(args, 'chairman', null)),
+    ) as Dict;
     const advisor_plans = _build_advisor_plans(ai_cfg, REPO_ROOT);
     const explicit_overrides = _parse_model_overrides(_getattr<string[] | null>(args, 'model', null));
     const skipped: Dict[] = [];
@@ -1978,6 +2027,21 @@ function cmd_run(
         _isDict(ai_cfg) &&
         _isDict(ai_cfg['stance_tally']) &&
         (ai_cfg['stance_tally'] as Dict)['enabled'] === true;
+    // road-to-council-blind-review Phase 1 (Ü2): five orthogonal stances
+    // rotated deterministically over the config-ordered member list; the
+    // outsider seat additionally drops project_context. Default off.
+    const stances_on = args.stances === true;
+    const stance_assignment = stances_on ? assign_stances(members.map((m) => m.name), question.user_prompt) : null;
+    const member_prompt_suffix = stance_assignment
+        ? new Map(Array.from(stance_assignment.entries()).map(([name, s]) => [name, s.prompt]))
+        : null;
+    const no_project_context_members = stance_assignment
+        ? new Set(
+              Array.from(stance_assignment.entries())
+                  .filter(([, s]) => s.name === OUTSIDER_STANCE_NAME)
+                  .map(([name]) => name),
+          )
+        : null;
     const stance_repairs: CouncilResponse[] = [];
     const responses = consult(members, question, budget, {
         table,
@@ -1986,6 +2050,8 @@ function cmd_run(
         rounds,
         advisor_plans,
         stance_tally: stance_tally_on,
+        member_prompt_suffix,
+        no_project_context_members,
         // Interactive one-line confirm (cmd_run has no --auto-continue); the
         // repaired-call cost is collected so cost_usd_actual stays honest.
         on_stance_repair: stance_tally_on
@@ -2009,6 +2075,22 @@ function cmd_run(
     );
     const consensus = _maybe_run_consensus(ai_cfg, question, members, responses, budget, table, project, args);
     const chairman = _maybe_run_chairman(ai_cfg, question, members, responses, budget, table, project, args);
+    // road-to-council-blind-review Phase 1 (Ü1), host-path only: when no
+    // member chairman ran (mode === 'host'), the blind mapping is computed
+    // here (seeded from the question text) and persisted for a later
+    // `council:render` to blind the response headers. The member-chairman
+    // path (below, inside `_maybe_run_chairman`) blinds inline instead.
+    const blind_chairman_on = args.blind_chairman === true;
+    const chairman_fields_on = args.chairman_fields === true;
+    let blind_review_map: Map<string, string> | null = null;
+    if (blind_chairman_on && chairman === null) {
+        const pairs: Array<[string, string]> = responses
+            .filter((r) => r.error === null && r.text.trim() !== '')
+            .map((r) => [`${r.provider}:${r.model}`, r.text]);
+        if (pairs.length > 0) {
+            blind_review_map = build_blind_labels(question.user_prompt, pairs).label_to_source;
+        }
+    }
     const estimated_total = estimates.reduce((acc, e) => acc + _total_usd(e), 0.0);
     let actual_total = 0.0;
     const all_responses: CouncilResponse[] = [...responses];
@@ -2043,6 +2125,9 @@ function cmd_run(
         cost_usd_estimated: _pyRound(estimated_total, 6),
         cost_usd_actual: _pyRound(actual_total, 6),
         stance_tally: stance_tally_on,
+        stances: stances_on,
+        blind_chairman: blind_chairman_on,
+        chairman_fields: chairman_fields_on,
         responses: _serialise_responses(responses),
     };
     if (peer_review !== null) {
@@ -2054,6 +2139,9 @@ function cmd_run(
     if (chairman !== null) {
         payload['chairman'] = { member: chairman.member, annotation: chairman.annotation, text: chairman.text };
     }
+    if (blind_review_map !== null) {
+        payload['blind_review_map'] = _mapToObject(blind_review_map);
+    }
     const out_path = _validate_council_output_path(args.output as string, {
         kind: 'responses',
         subcommand: 'run',
@@ -2452,12 +2540,18 @@ function cmd_render(args: Args): number {
     }
     const consensus = _deserialise_consensus(payload['consensus'] as Dict);
     const peer_review = _deserialise_peer_review(payload['peer_review'] as Dict);
+    // road-to-council-blind-review Phase 1: a persisted `blind_review_map`
+    // means `--blind-chairman` ran on the host path (no member chairman) —
+    // blind the response headers + append the de-anonymization map.
+    const blind_map_raw = payload['blind_review_map'];
     const body = render(_deserialise_responses(items), {
         mode: mode ?? null,
         prose_synthesis: prose,
         consensus,
         peer_review,
         stance_tally: payload['stance_tally'] === true,
+        chairman_fields: payload['chairman_fields'] === true,
+        blind: _isDict(blind_map_raw) ? { label_to_source: _objToMap(blind_map_raw as Dict) } : null,
         chairman: _isDict(payload['chairman'])
             ? (payload['chairman'] as { member: string | null; annotation: string; text: string | null })
             : null,
@@ -2672,6 +2766,11 @@ interface Args {
     auto_continue: boolean;
     restate?: boolean;
     continue_as_debate: string | null;
+    // road-to-council-blind-review Phase 1 — flag-gated, default-off (`run` only)
+    chairman: string | null;
+    blind_chairman: boolean;
+    stances: boolean;
+    chairman_fields: boolean;
     // render / replay
     responses?: string | null;
     include_member_arguments: boolean | null;
@@ -2726,6 +2825,15 @@ function _defaultArgs(): Args {
         prose_synthesis: null,
         auto_continue: false,
         continue_as_debate: null,
+        chairman: null,
+        // Ü1 ADOPTED (road-to-council-blind-review Phase 3, 2026-07-28):
+        // blind synthesis is the default — 0/10 + 0/10 pre-registered
+        // degradation triggers on the n=10 A/B; opt out per-invocation
+        // with --no-blind-chairman (the audit artifact always keeps the
+        // post-verdict de-anonymization map either way).
+        blind_chairman: true,
+        stances: false,
+        chairman_fields: false,
         responses: null,
         include_member_arguments: null,
         low_impact_stats: false,
@@ -2788,6 +2896,12 @@ function _specsFor(cmd: string): { positionals: string[]; opts: OptSpec[]; requi
                     { flag: '--single', takesValue: false, apply: (o) => (o.single = true) },
                     { flag: '--prose-synthesis', takesValue: false, apply: (o) => (o.prose_synthesis = true) },
                     { flag: '--no-prose-synthesis', takesValue: false, apply: (o) => (o.prose_synthesis = false) },
+                    // road-to-council-blind-review Phase 1 — flag-gated, default-off.
+                    { flag: '--chairman', takesValue: true, apply: (o, v) => (o.chairman = v) },
+                    { flag: '--blind-chairman', takesValue: false, apply: (o) => (o.blind_chairman = true) },
+                    { flag: '--no-blind-chairman', takesValue: false, apply: (o) => (o.blind_chairman = false) },
+                    { flag: '--stances', takesValue: false, apply: (o) => (o.stances = true) },
+                    { flag: '--chairman-fields', takesValue: false, apply: (o) => (o.chairman_fields = true) },
                 ],
                 requiredOpts: ['--output'],
             };

package/src/scripts/detect_target_license.ts

@@ -0,0 +1,203 @@
+#!/usr/bin/env node
+/**
+ * detect_target_license — deterministic SPDX license detection + derived
+ * borrow-compatibility policy (road-to-provenance-and-license-governance
+ * S1.2). CLI wrapper — the actual detection/matrix/invalidation/override
+ * logic lives in `./_lib/detect_target_license.ts` (that module does the
+ * filesystem I/O too; detection is inherently a file-read job). This file
+ * owns only argv parsing, stdout/stderr shaping, schema validation, and the
+ * `--write` side effect.
+ *
+ * Usage:
+ *   npx tsx src/scripts/detect_target_license.ts [repo-path] [--write] [--json]
+ *   npx tsx src/scripts/detect_target_license.ts --help
+ *
+ * Default (no --write): report only, nothing written to disk.
+ * --write   create/update license-policy.yaml in the target repo, unless
+ *           escalation is required or the existing file is derived_from: manual.
+ *
+ * Exit codes: 0 success (including the warn-only "nothing detectable"
+ * path); 1 usage/IO/schema error; 2 escalation required — a human decision,
+ * never an auto-pick (source disagreement, or a divergent workspace license).
+ */
+
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import { pathToFileURL } from 'node:url';
+
+import { parse as parseYaml, stringify as stringifyYaml } from 'yaml';
+
+import {
+    buildLicensePolicyDocument,
+    checkInvalidation,
+    detectTargetLicense,
+    mergeOverride,
+    type LicensePolicyDoc,
+} from './_lib/detect_target_license.js';
+import { load_schema, validate, type YamlValue } from './validate_frontmatter.js';
+
+export const POLICY_FILENAME = 'license-policy.yaml';
+
+const USAGE = `usage: detect_target_license [repo-path] [--write] [--json] [-h|--help]
+
+Deterministic SPDX license detection + derived borrow-compatibility policy.
+
+  repo-path   target repo root (default: .)
+  --write     write/update ${POLICY_FILENAME} in the target repo (skipped on
+              escalation, or when the existing file is derived_from: manual)
+  --json      machine-readable report on stdout
+
+Exit codes: 0 success, 1 usage/IO/schema error, 2 escalation required.
+`;
+
+export interface CliArgs {
+    repoPath: string;
+    write: boolean;
+    json: boolean;
+    help: boolean;
+}
+
+export function parseArgs(argv: string[]): CliArgs {
+    let repoPath = '.';
+    let write = false;
+    let json = false;
+    let help = false;
+    for (const a of argv) {
+        if (a === '-h' || a === '--help') help = true;
+        else if (a === '--write') write = true;
+        else if (a === '--json') json = true;
+        else if (!a.startsWith('--')) repoPath = a;
+    }
+    return { repoPath, write, json, help };
+}
+
+export function serializeLicensePolicy(doc: LicensePolicyDoc): string {
+    return stringifyYaml(doc, { indent: 2 });
+}
+
+export function parseLicensePolicyFile(text: string): LicensePolicyDoc {
+    return parseYaml(text) as LicensePolicyDoc; // caller schema-validates before trusting this
+}
+
+export function main(argv: string[] = process.argv.slice(2)): number {
+    const args = parseArgs(argv);
+    if (args.help) {
+        process.stdout.write(USAGE);
+        return 0;
+    }
+
+    const repoRoot = path.resolve(args.repoPath);
+    if (!fs.existsSync(repoRoot) || !fs.statSync(repoRoot).isDirectory()) {
+        process.stderr.write(`detect_target_license: error: not a directory: ${repoRoot}\n`);
+        return 1;
+    }
+
+    const detection = detectTargetLicense(repoRoot);
+
+    if (detection.escalate) {
+        if (args.json) {
+            process.stdout.write(JSON.stringify({
+                escalate: true,
+                reason: detection.escalateReason,
+                spdx_id: detection.spdxId,
+                warnings: detection.warnings,
+            }) + '\n');
+        } else {
+            process.stdout.write(`detect_target_license: ESCALATE — ${detection.escalateReason ?? 'unspecified'}\n`);
+            for (const w of detection.warnings) process.stdout.write(`  warning: ${w}\n`);
+            process.stdout.write(`No ${POLICY_FILENAME} written. This needs a human decision, never an auto-pick.\n`);
+        }
+        return 2;
+    }
+
+    const nowIso = new Date().toISOString();
+    const freshDoc = buildLicensePolicyDocument(detection, nowIso);
+    const policyPath = path.join(repoRoot, POLICY_FILENAME);
+
+    let finalDoc: LicensePolicyDoc = freshDoc;
+    let rejectedOverrides: string[] = [];
+    let action: 'derived' | 'kept-manual' | 're-derived' | 'merged' | 'unchanged' = 'derived';
+
+    if (fs.existsSync(policyPath)) {
+        let existing: LicensePolicyDoc | null = null;
+        try {
+            existing = parseLicensePolicyFile(fs.readFileSync(policyPath, 'utf8'));
+        } catch (err) {
+            process.stderr.write(
+                `detect_target_license: warning: failed to parse existing ${POLICY_FILENAME}, ` +
+                `treating as absent: ${err instanceof Error ? err.message : String(err)}\n`,
+            );
+        }
+
+        if (existing) {
+            const invalidation = checkInvalidation(existing, detection);
+            if (invalidation.action === 'escalate') {
+                if (args.json) {
+                    process.stdout.write(JSON.stringify({ escalate: true, reason: invalidation.reason }) + '\n');
+                } else {
+                    process.stdout.write(`detect_target_license: ESCALATE — ${invalidation.reason}\n`);
+                    process.stdout.write(`${POLICY_FILENAME} left unchanged. This needs a human decision, never an auto-pick.\n`);
+                }
+                return 2;
+            }
+            if (existing.derived_from === 'manual') {
+                finalDoc = existing;
+                action = 'kept-manual';
+            } else if (invalidation.action === 're-derive') {
+                finalDoc = freshDoc;
+                action = 're-derived';
+            } else {
+                const merge = mergeOverride(existing, freshDoc);
+                finalDoc = merge.doc;
+                rejectedOverrides = merge.rejectedOverrides;
+                action = rejectedOverrides.length > 0 || JSON.stringify(merge.doc.policy) !== JSON.stringify(existing.policy)
+                    ? 'merged'
+                    : 'unchanged';
+            }
+        }
+    }
+
+    if (action !== 'kept-manual') {
+        const schema = load_schema('license-policy');
+        const errors = validate(finalDoc as unknown as YamlValue, schema);
+        if (errors.length > 0) {
+            process.stderr.write('detect_target_license: refused — derived policy fails schema validation:\n');
+            for (const e of errors) process.stderr.write(`  - ${e.format()}\n`);
+            return 1;
+        }
+    }
+
+    const willWrite = args.write && action !== 'kept-manual';
+    if (willWrite) {
+        fs.writeFileSync(policyPath, serializeLicensePolicy(finalDoc), 'utf8');
+    }
+
+    if (args.json) {
+        process.stdout.write(JSON.stringify({
+            escalate: false,
+            action,
+            wrote: willWrite,
+            policy_path: policyPath,
+            doc: finalDoc,
+            rejected_overrides: rejectedOverrides,
+        }) + '\n');
+    } else {
+        process.stdout.write(
+            `detect_target_license: detected ${detection.spdxId ?? '(none)'} -> ` +
+            `target_class=${detection.targetClass} workspace_scope=${detection.workspaceScope}\n`,
+        );
+        for (const w of detection.warnings) process.stdout.write(`  warning: ${w}\n`);
+        for (const r of rejectedOverrides) process.stdout.write(`  rejected override: ${r}\n`);
+        process.stdout.write(
+            willWrite
+                ? `wrote ${policyPath} (${action})\n`
+                : `(dry run — pass --write to update ${policyPath}; action would be: ${action})\n`,
+        );
+    }
+    return 0;
+}
+
+if (process.argv[1] !== undefined) {
+    const invokedUrl = pathToFileURL(path.resolve(process.argv[1])).href;
+    if (import.meta.url === invokedUrl) process.exit(main());
+}

package/src/scripts/evaluator_umbrella.sh

@@ -0,0 +1,145 @@
+#!/usr/bin/env bash
+# evaluator_umbrella.sh — the external evaluator's "first five minutes" as
+# ONE suite over the PACKED artifact (road-to-credible-install Phase 6).
+#
+# Runs in a clean container (see .github/workflows/evaluator-umbrella.yml)
+# on every release PR and nightly — never against the checkout's working
+# tree: every check below exercises the tarball `npm pack` produces or the
+# clean consumer install made from it.
+#
+# Checks (each individually red-testable via its own gate/tests):
+#   1. audit gate           npm audit --omit=dev --audit-level=high
+#   2. tarball integrity    publint + prepack script-target assertion
+#                           (lifecycle gate: tests/scripts/prepack_lifecycle_check.test.ts)
+#   3. headless install     npm install --omit=dev <tarball> in a scratch project
+#   4. staleness lints      lint_pre_migration_refs (+ consumer-internal refs
+#                           lint when present)
+#   5. hook-latency bench   bench_hook_latency --gate (budget + regression)
+#   6. cold-start budgets   CLI --version, mcp-server boot-to-initialize
+#   7. size budgets         unpacked, node_modules, dep count (>10% regression
+#                           fails even under budget)
+#   8. surface budgets      CLI registry command count, MCP tools/list count
+#                           (frozen post-Phase-3 values, ADR-132: 19)
+#   9. npx smoke            first-touch paths (--version, --help, mcp initialize)
+#   6-8 feed src/scripts/check_evaluator_budgets.ts against
+#   src/config/evaluator-budgets.json (owner + review date lint-enforced).
+#
+# Usage: bash src/scripts/evaluator_umbrella.sh [--skip-pack <tarball>]
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$ROOT"
+
+TARBALL=""
+if [[ "${1:-}" == "--skip-pack" && -n "${2:-}" ]]; then
+  TARBALL="$2"
+fi
+
+step() { printf '\n=== umbrella: %s ===\n' "$1"; }
+
+# ---------------------------------------------------------------- 1. audit
+step "npm audit (runtime deps, high+)"
+npm audit --omit=dev --audit-level=high
+
+# ------------------------------------------------------------- 2. pack + shape
+if [[ -z "$TARBALL" ]]; then
+  step "npm pack (prepack chain: build + discovery + manifest + prepack-check)"
+  TARBALL="$(npm pack --silent | tail -1)"
+else
+  step "using pre-packed tarball: $TARBALL"
+fi
+[[ -f "$TARBALL" ]] || { echo "❌ tarball missing: $TARBALL" >&2; exit 1; }
+
+step "publint (published-package shape)"
+npm run lint:publint --silent
+
+# `--ignore-scripts` is NOT honoured for `prepare` by every npm version: on the
+# node-20 container npm still runs it, and this repo's prepare installs the git
+# hooks and prints a banner — straight into the `--json` stdout we parse. Slice
+# from the first `[` so any lifecycle banner ahead of the payload is tolerated
+# instead of turning into `SyntaxError: Unexpected token '✅'`.
+UNPACKED_BYTES="$(npm pack --dry-run --json --ignore-scripts --silent 2>/dev/null \
+  | node -e 'let b="";process.stdin.on("data",c=>b+=c).on("end",()=>{const i=b.indexOf("[");const j=JSON.parse(i>=0?b.slice(i):b);console.log(j[0].unpackedSize)})')"
+
+# ------------------------------------------------------- 3. headless install
+step "headless consumer install from the tarball"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+( cd "$WORK" && npm init -y >/dev/null 2>&1 \
+  && npm install --omit=dev --no-audit --no-fund --silent "$ROOT/$TARBALL" )
+PKG_DIR="$WORK/node_modules/@event4u/agent-config"
+[[ -d "$PKG_DIR" ]] || { echo "❌ install did not produce the package dir" >&2; exit 1; }
+
+# --------------------------------------------------------- 4. staleness lints
+step "staleness lints (pre-migration refs)"
+./scripts-run src/scripts/lint_pre_migration_refs
+if [[ -f src/scripts/lint_consumer_internal_refs.ts ]]; then
+  ./scripts-run src/scripts/lint_consumer_internal_refs
+fi
+
+# ------------------------------------------------------ 5. hook-latency bench
+step "hook-latency bench (pre-registered budget gate)"
+./scripts-run src/scripts/bench_hook_latency --gate
+
+# ----------------------------------------------- 6-8. measured budget metrics
+step "cold-start + size + surface measurements"
+MEASURE_TMP="$WORK/measurements.json"
+CLI_BIN="$WORK/node_modules/.bin/agent-config"
+
+CLI_MS="$(node -e '
+const {execFileSync}=require("node:child_process");
+const t=Date.now();
+execFileSync(process.argv[1],["--version"],{stdio:"ignore"});
+console.log(Date.now()-t);' "$CLI_BIN")"
+
+MCP_MS="$(node -e '
+const {spawnSync}=require("node:child_process");
+const t=Date.now();
+const p=spawnSync("node",[process.argv[1]+"/dist/mcp/server.mjs"],{input:JSON.stringify({jsonrpc:"2.0",id:1,method:"initialize",params:{protocolVersion:"2024-11-05",capabilities:{},clientInfo:{name:"p",version:"0"}}})+"\n",encoding:"utf-8",timeout:30000});
+if(!(p.stdout||"").includes("serverInfo")){console.error(p.stderr);process.exit(1)}
+console.log(Date.now()-t);' "$PKG_DIR")"
+
+NM_MB="$(du -sm "$WORK/node_modules" | awk '{print $1}')"
+DEP_COUNT="$(node -e '
+const j=require(process.argv[1]);
+console.log(Object.keys(j.packages||{}).length-1);' "$WORK/package-lock.json")"
+
+CLI_COUNT="$(node -e '
+const fs=require("node:fs");
+const t=fs.readFileSync(process.argv[1],"utf-8");
+console.log((t.match(/\{ name: /g)||[]).length);' "$ROOT/src/cli/registry.ts")"
+
+MCP_TOOLS="$(node -e '
+const {spawnSync}=require("node:child_process");
+const reqs=[{jsonrpc:"2.0",id:1,method:"initialize",params:{protocolVersion:"2024-11-05",capabilities:{},clientInfo:{name:"p",version:"0"}}},{jsonrpc:"2.0",method:"notifications/initialized"},{jsonrpc:"2.0",id:2,method:"tools/list"}];
+const p=spawnSync("node",[process.argv[1]+"/dist/mcp/server.mjs"],{input:reqs.map(r=>JSON.stringify(r)).join("\n")+"\n",encoding:"utf-8",timeout:30000});
+const lines=(p.stdout||"").trim().split("\n").map(l=>{try{return JSON.parse(l)}catch{return null}}).filter(Boolean);
+const resp=lines.find(l=>l.id===2);
+if(!resp||!resp.result||!Array.isArray(resp.result.tools)){console.error("no tools/list result");process.exit(1)}
+console.log(resp.result.tools.length);' "$PKG_DIR")"
+
+node -e '
+const fs=require("node:fs");
+const m={
+  unpacked_size_mb: Number(process.argv[2])/1e6,
+  node_modules_mb: Number(process.argv[3]),
+  runtime_dep_count: Number(process.argv[4]),
+  cli_version_cold_ms: Number(process.argv[5]),
+  mcp_boot_to_initialize_ms: Number(process.argv[6]),
+  cli_help_command_count: Number(process.argv[7]),
+  mcp_public_tool_count: Number(process.argv[8]),
+};
+fs.writeFileSync(process.argv[1], JSON.stringify(m, null, 2));
+console.log(JSON.stringify(m, null, 1));' \
+  "$MEASURE_TMP" "$UNPACKED_BYTES" "$NM_MB" "$DEP_COUNT" "$CLI_MS" "$MCP_MS" "$CLI_COUNT" "$MCP_TOOLS"
+
+step "budget gate (absolute + >regression_pct creep)"
+./scripts-run src/scripts/check_evaluator_budgets --measurements "$MEASURE_TMP"
+
+# ------------------------------------------------------------- 9. npx smoke
+step "first-touch smoke (installed bin)"
+"$CLI_BIN" --version >/dev/null
+"$CLI_BIN" --help >/dev/null
+
+echo
+echo "✅  evaluator umbrella green — the first five minutes hold on the packed artifact."

package/src/scripts/external_sources_denylist.json

@@ -57,7 +57,9 @@
     "\\bskillui\\b",
     "\\bgraphify\\b",
     "graphify-labs",
-    "penpax"
+    "penpax",
+    "\\bagent-reach\\b",
+    "panniantong"
   ],
   "skip_paths": [
     "src/scripts/check_no_external_sources.py",
@@ -89,15 +91,17 @@
     "agents/roadmaps/archive/road-to-final-state-and-market-readiness.md",
     "agents/roadmaps/archive/road-to-image-brand-typography.md",
     "agents/roadmaps/archive/road-to-subagent-value-realization.md",
-    "agents/roadmap-assets/road-to-image-brand-typography.assets.md"
+    "agents/roadmap-assets/road-to-image-brand-typography.assets.md",
+    "CREDITS.md",
+    "agents/roadmaps/archive/road-to-ecosystem-harvest-index.md"
   ],
   "skip_reason": {
-    "vendored_cluster": "design-intelligence/corpus-grounding/design-tokens/react-shadcn-ui/tailwind-engineer + cost/*.mjs ship vendored Apache/MIT code; their upstream attribution is license-required and must NOT be stripped (ADR-061).",
+    "vendored_cluster": "design-intelligence/corpus-grounding/design-tokens/react-shadcn-ui/tailwind-engineer + cost/*.mjs ship vendored Apache/MIT code; their upstream attribution is license-required and must NOT be stripped (ADR-061). CREDITS.md is the repo-level attribution file FOR that same cluster — it names the upstreams because the licenses require it, which is the one case source-confidentiality explicitly exempts. It was missing here, so the gate flagged the very attribution the rule mandates.",
     "recommendation_docs": "mcp.md, mcp-registries.md, DISTRIBUTION_CHECKLIST.md, ADR-086, marketing-copy.yml reference registries (Smithery/Glama) as recommendations/distribution targets — allowed.",
     "generated_bundles": "content.json* bundle the vendored-cluster skill bodies above.",
     "self": "the linter + its denylist + the source-confidentiality rule necessarily contain the deny tokens as data.",
     "validate_safe_paths": "telegraph-derived (MIT) ported code; upstream attribution is license-required.",
-    "archived_roadmaps_newly_tracked": "Archived roadmaps (and their assets) that predate the source-confidentiality sweep became tracked via the 2026-07 agents-dir gitignore hygiene change. They are frozen historical records; the claudekit mentions are ADR-061 license-attribution work and the remaining names are historical planning references. New roadmaps stay subject to the gate."
+    "archived_roadmaps_newly_tracked": "Archived roadmaps (and their assets) that predate the source-confidentiality sweep became tracked via the 2026-07 agents-dir gitignore hygiene change. They are frozen historical records; the claudekit mentions are ADR-061 license-attribution work and the remaining names are historical planning references. New roadmaps stay subject to the gate. road-to-ecosystem-harvest-index.md is the sharpest case: the flagged line is its own acceptance criterion, 'No harvest-family artifact names a real external source' — it quotes the names in order to assert their absence, so the gate fires on the sentence that records the cleanup. road-to-opt-retrieval-and-memory.md was REMOVED from this list on 2026-07-28 (road-to-native-code-intelligence Phase 6): its two clear-text mentions named a gitignored borrow-catalogue inbox file, which carries no historical value that a neutral descriptor loses — so the mentions were rewritten and the exception retired rather than kept. Retiring an exception is preferred over holding one whenever the named token is not license-required attribution and not load-bearing for the record."
   },
   "_excluded_tokens": {
     "agent-os": "collides with our own coined 'Agent-OS primitive' term (universal-skills) — too FP-prone to auto-deny; scrub external 'agent-os' mentions by review.",