@event4u/agent-config 14.11.0 → 14.13.0

package/src/scripts/lint_ticket_buildable.ts

@@ -609,6 +609,22 @@ function _cycle(graph: Map<string, string[]>): string[] | null {
 // iterdir / glob helpers (pathlib-faithful sort)
 // ---------------------------------------------------------------------------
 
+/**
+ * Directories under `agents/tickets/` that are NOT ticket bundles.
+ *
+ * `archive` was the original member and was tested inline. `symptoms` joins it
+ * because it is a different artefact class that happens to live under the same
+ * root: one markdown file per operator symptom report, plus a README and a
+ * template, established so a production symptom had somewhere in the tree to
+ * land. It has no tickets, no dependency graph and no Definition-of-Ready, so
+ * every check below is meaningless there — and demanding a `manifest.yml` from
+ * it is the gate asserting a contract that directory never claimed.
+ *
+ * A named set rather than a second `||`: the next non-bundle directory should
+ * cost one line here instead of another inline special case.
+ */
+const NON_BUNDLE_DIRS: ReadonlySet<string> = new Set(['archive', 'symptoms']);
+
 /** sorted(TICKETS_ROOT.iterdir()) — all entries (files + dirs), name-sorted. */
 function _iterdirSorted(root: string): string[] {
     let names: string[];
@@ -707,7 +723,7 @@ function lint(): number {
     const bundle_ids = new Set<string>();
 
     for (const bundle of _iterdirSorted(TICKETS_ROOT)) {
-        if (!_isDir(bundle) || path.basename(bundle) === 'archive') {
+        if (!_isDir(bundle) || NON_BUNDLE_DIRS.has(path.basename(bundle))) {
             continue;
         }
 

package/src/scripts/mcp_server/consumer_tool_catalog.json

@@ -1,6 +1,6 @@
 {
   "schema_version": 1,
-  "description": "Source-of-truth catalog of consumer-relevant MCP tools. Read by the stdio server (scripts/mcp_server/) and packed into the Cloud Worker bundle (internal/workers/mcp/). The 'implemented_on' field lists transports where the real handler is wired. Entries marked '[stub — implemented on demand]' are DOCUMENTATION of the on-demand backlog: per ADR-132 they are NOT registered on the stdio tools/list (the wire carries implemented tools only); the 'not_implemented' envelope contract (docs/contracts/mcp-tool-stub-envelope.md) remains for Worker parity. Generated by `src/scripts/build_mcp_catalog.ts` — never hand-edit this file.",
+  "description": "Source-of-truth catalog of consumer-relevant MCP tools. Read by the stdio server (scripts/mcp_server/) and packed into the Cloud Worker bundle (internal/workers/mcp/). The 'implemented_on' field lists transports where the real handler is wired. Entries marked '[stub — implemented on demand]' are DOCUMENTATION of the on-demand backlog: per ADR-132 they are NOT registered on the stdio tools/list (the wire carries implemented tools only); the 'not_implemented' envelope contract (docs/contracts/mcp-tool-stub-envelope.md) remains for Worker parity. Generated by `build_mcp_catalog` — never hand-edit this file.",
   "install_hint_stdio": "npx -y @event4u/agent-config mcp-server",
   "tools": [
     {

package/src/scripts/mcp_server/tool_catalog_source.ts

@@ -33,7 +33,7 @@ export const CATALOG_DESCRIPTION =
     'backlog: per ADR-132 they are NOT registered on the stdio tools/list ' +
     "(the wire carries implemented tools only); the 'not_implemented' " +
     'envelope contract (docs/contracts/mcp-tool-stub-envelope.md) remains ' +
-    'for Worker parity. Generated by `src/scripts/build_mcp_catalog.ts` — ' +
+    'for Worker parity. Generated by `build_mcp_catalog` — ' +
     'never hand-edit this file.';
 
 export interface StubToolSource {

package/src/scripts/measure_docs_dead_links.ts

@@ -0,0 +1,348 @@
+#!/usr/bin/env tsx
+/**
+ * Dead relative links under `docs/`, split by whether the file SHIPS.
+ *
+ * `road-to-contract-review-deadlines` Phase 4.1. `check_references.ts` scans
+ * `dist/agent-src` and `agents` only, so `docs/` is green by exclusion rather
+ * than by being correct — the run reports thousands of `excluded_directory`
+ * skips and no other gate does general link validation there.
+ *
+ * The split is the point, not the total. A dead link in `docs/guidelines/` is
+ * shipped to consumers and is the only half that reaches anyone; a dead link in
+ * an internal analysis note costs a maintainer one confused minute. Reporting
+ * one number for both invites either an unreviewable 487-line sweep or nothing.
+ *
+ * A `.py` target is broken out separately because it is a MIGRATION LEFTOVER
+ * with a mechanical fix — the TypeScript port renamed the file, and the pointer
+ * did not follow. It is a class, not 152 independent defects.
+ *
+ * MEASUREMENT ONLY. This is deliberately not a `lint_*` / `check_*` gate: the
+ * scope decision (does `docs/` enter `check_references`?) is step 4.3 and is a
+ * maintainer call. A gate landed before that decision would pre-empt it.
+ *
+ * Usage:
+ *     ./scripts-run src/scripts/measure_docs_dead_links
+ *     ./scripts-run src/scripts/measure_docs_dead_links --format json
+ *     ./scripts-run src/scripts/measure_docs_dead_links --list shipped
+ */
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import process from 'node:process';
+import { fileURLToPath, pathToFileURL } from 'node:url';
+
+const _HERE = fileURLToPath(import.meta.url);
+const REPO_ROOT = path.resolve(path.dirname(_HERE), '..', '..');
+
+/** Roots under `docs/` that `package.json:files[]` actually publishes. */
+export const SHIPPED_DOCS_PREFIXES = [
+    'docs/guidelines/',
+    'docs/contracts/persona-schema.md',
+    'docs/contracts/provider-lifecycle.md',
+    'docs/contracts/settings-classes.md',
+] as const;
+
+/** Markdown inline links: `[text](target)`. Reference-style links are out of scope. */
+const LINK_RE = /\[[^\]]*\]\(([^)\s]+)(?:\s+"[^"]*")?\)/g;
+
+export type DeadClass = 'shipped' | 'internal';
+
+export interface DeadLink {
+    file: string;
+    target: string;
+    resolved: string;
+    cls: DeadClass;
+    pyTarget: boolean;
+}
+
+export interface Measurement {
+    filesScanned: number;
+    linksScanned: number;
+    dead: DeadLink[];
+}
+
+function walkMd(root: string, rel = ''): string[] {
+    const out: string[] = [];
+    const abs = path.join(root, rel);
+    for (const ent of fs.readdirSync(abs, { withFileTypes: true }).sort((a, b) => (a.name < b.name ? -1 : 1))) {
+        const r = rel === '' ? ent.name : `${rel}/${ent.name}`;
+        if (ent.isDirectory()) out.push(...walkMd(root, r));
+        else if (ent.name.endsWith('.md')) out.push(r);
+    }
+    return out;
+}
+
+function isShipped(file: string): boolean {
+    return SHIPPED_DOCS_PREFIXES.some((p) => (p.endsWith('/') ? file.startsWith(p) : file === p));
+}
+
+/**
+ * True for a link this measurement can decide. Anchors, URLs, mailto: and
+ * absolute paths are not relative file references and are skipped rather than
+ * counted as alive — counting an undecidable target as alive is the same
+ * false-green this whole roadmap is about.
+ */
+function isRelativeFileLink(target: string): boolean {
+    if (target === '' || target.startsWith('#')) return false;
+    // Prose placeholders, not links: a bare ellipsis inside a code-ish span, or
+    // an angle-bracket variable like `../<peer>/SKILL.md`. Counting them as dead
+    // would put un-fixable rows in a count whose whole purpose is to reach zero.
+    if (/^\.+$/.test(target)) return false;
+    if (target.includes('<') || target.includes('>')) return false;
+    if (/^[a-z][a-z0-9+.-]*:/i.test(target)) return false; // http:, https:, mailto:, …
+    if (target.startsWith('/')) return false;
+    return true;
+}
+
+export function measure(repoRoot: string = REPO_ROOT): Measurement {
+    const docsRoot = path.join(repoRoot, 'docs');
+    const files = walkMd(docsRoot);
+    const dead: DeadLink[] = [];
+    let linksScanned = 0;
+
+    for (const rel of files) {
+        const abs = path.join(docsRoot, rel);
+        const text = fs.readFileSync(abs, 'utf8');
+        LINK_RE.lastIndex = 0;
+        let m: RegExpExecArray | null;
+        while ((m = LINK_RE.exec(text)) !== null) {
+            const raw = m[1]!;
+            if (!isRelativeFileLink(raw)) continue;
+            linksScanned += 1;
+            const target = raw.split('#')[0]!;
+            if (target === '') continue; // pure anchor after the split
+            const resolved = path.normalize(path.join(path.dirname(abs), target));
+            if (fs.existsSync(resolved)) continue;
+            const file = `docs/${rel}`;
+            dead.push({
+                file,
+                target: raw,
+                resolved: path.relative(repoRoot, resolved),
+                cls: isShipped(file) ? 'shipped' : 'internal',
+                pyTarget: target.endsWith('.py'),
+            });
+        }
+    }
+    return { filesScanned: files.length, linksScanned, dead };
+}
+
+
+/**
+ * Repair a dead target by finding the ONE file in the tree it can only mean.
+ *
+ * The tail heuristic, and why it is safe: nearly every dead link here points at
+ * a container that MOVED — the retired uncondensed source tree after ADR-051
+ * flattened the source tree, or `../../rules/x.md` written from the wrong depth.
+ * The container changed; the tail did not. So the tail from a known segment
+ * onward (`rules/x.md`, `skills/x/SKILL.md`) still names the file uniquely.
+ *
+ * **Exactly one match, or no rewrite.** Ambiguity is reported, never guessed:
+ * a link silently repointed at the wrong one of two candidates is worse than a
+ * dead link, because a dead link is visible and a wrong one is not.
+ */
+const TAIL_SEGMENTS = ['rules/', 'skills/', 'commands/', 'contexts/', 'templates/', 'personas/', 'guidelines/', 'contracts/', 'scripts/', 'schemas/', 'tests/', 'golden/', 'decisions/'] as const;
+
+function repoFileIndex(repoRoot: string): string[] {
+    const out: string[] = [];
+    const skip = new Set(['node_modules', '.git', 'dist', '.augment', '.claude', '.cursor', '.clinerules']);
+    const walk = (rel: string): void => {
+        for (const ent of fs.readdirSync(path.join(repoRoot, rel), { withFileTypes: true })) {
+            if (skip.has(ent.name)) continue;
+            const r = rel === '' ? ent.name : `${rel}/${ent.name}`;
+            if (ent.isDirectory()) walk(r);
+            else out.push(r);
+        }
+    };
+    walk('');
+    return out;
+}
+
+export function resolveTail(target: string, index: readonly string[]): string | null {
+    const clean0 = target.split('#')[0]!;
+    // The migration class, handled first because it is a RENAME rather than a
+    // move: the TypeScript port changed the extension and the pointer did not
+    // follow. `scripts/x.py` -> `scripts/x.ts` is mechanical and unambiguous
+    // when the `.ts` file exists; when it does not, the file was deleted rather
+    // than ported and the link falls through to the tail search below.
+    const candidates = clean0.endsWith('.py')
+        ? [
+              `${clean0.slice(0, -3)}.ts`,
+              // pytest's `test_x.py` became vitest's `x.test.ts` — a rename with
+              // a different SHAPE, not just a different extension, so it needs
+              // its own candidate rather than an extension swap.
+              clean0.replace(/(^|\/)test_([^/]+)\.py$/, '$1$2.test.ts'),
+              clean0,
+          ]
+        : [clean0];
+    for (const clean of candidates) {
+        const hit = resolveOne(clean, index);
+        if (hit !== null) return hit;
+    }
+    return null;
+}
+
+function resolveOne(clean: string, index: readonly string[]): string | null {
+    for (const seg of TAIL_SEGMENTS) {
+        const at = clean.lastIndexOf(seg);
+        if (at === -1) continue;
+        const tail = clean.slice(at);
+        const hits = index.filter((f) => f === tail || f.endsWith(`/${tail}`));
+        if (hits.length === 1) return hits[0]!;
+        if (hits.length > 1) {
+            // Tie-break on the ONE axis this repository has already settled:
+            // `src/` is the single source of truth and every other tree is a
+            // projection or an override (CLAUDE.md § Source of truth). So a tie
+            // between `src/rules/x.md` and `agents/overrides/rules/x.md` is not
+            // a real ambiguity — a doc pointing at the rule means the source.
+            // Any tie this does NOT break stays unresolved.
+            // Prefer the candidate the ORIGINAL target already named. A link
+            // reading `../../../src/scripts/memory_status.py` ties between
+            // `src/scripts/memory_status.ts` and
+            // `src/agent-src/templates/scripts/memory_status.ts` on the tail
+            // alone — but only one of them is what the author wrote, once the
+            // `../` prefix is dropped.
+            const normalized = clean.replace(/^(?:\.\.\/)+/, '');
+            const exact = hits.filter((f) => f === normalized);
+            if (exact.length === 1) return exact[0]!;
+            const inSrc = hits.filter((f) => f.startsWith('src/'));
+            if (inSrc.length === 1) return inSrc[0]!;
+            return null;
+        }
+    }
+    // Last resort: the BASENAME alone, and only when the whole tree holds
+    // exactly one file with it. This is what catches a move that changed the
+    // directory as well as the name — `tests/test_condense.py` became
+    // `tests/scripts/condense.test.ts`, so neither the tail nor the extension
+    // swap alone reaches it. A basename with two homes stays unresolved: the
+    // uniqueness IS the safety property, not a convenience.
+    const base = clean.split('/').pop() ?? '';
+    if (base === '' || !base.includes('.')) return null;
+    const byBase = index.filter((f) => f.endsWith(`/${base}`) || f === base);
+    if (byBase.length === 1) return byBase[0]!;
+    const srcOrTests = byBase.filter((f) => f.startsWith('src/') || f.startsWith('tests/'));
+    if (srcOrTests.length === 1) return srcOrTests[0]!;
+    return null;
+}
+
+interface FixResult {
+    rewritten: number;
+    files: number;
+    unresolved: DeadLink[];
+}
+
+export function applyFix(repoRoot: string, dead: readonly DeadLink[]): FixResult {
+    const index = repoFileIndex(repoRoot);
+    const byFile = new Map<string, DeadLink[]>();
+    for (const d of dead) {
+        const l = byFile.get(d.file) ?? [];
+        l.push(d);
+        byFile.set(d.file, l);
+    }
+    const unresolved: DeadLink[] = [];
+    let rewritten = 0;
+    let filesTouched = 0;
+    for (const [file, allLinks] of byFile) {
+        // One rewrite per DISTINCT target: the replacement is global within the
+        // file, so a target appearing twice would otherwise report one success
+        // and one phantom "unresolved" for a link that is already fixed.
+        const seen = new Set<string>();
+        const links = allLinks.filter((d) => (seen.has(d.target) ? false : (seen.add(d.target), true)));
+        const abs = path.join(repoRoot, file);
+        let text = fs.readFileSync(abs, 'utf8');
+        let changed = false;
+        for (const d of links) {
+            const resolvedRel = resolveTail(d.target, index);
+            if (resolvedRel === null) {
+                unresolved.push(d);
+                continue;
+            }
+            const anchor = d.target.includes('#') ? `#${d.target.split('#').slice(1).join('#')}` : '';
+            let next = path.relative(path.dirname(abs), path.join(repoRoot, resolvedRel));
+            if (!next.startsWith('.')) next = `./${next}`;
+            const replacement = `${next}${anchor}`;
+            if (replacement === d.target) {
+                unresolved.push(d);
+                continue;
+            }
+            const needle = `](${d.target})`;
+            if (!text.includes(needle)) {
+                unresolved.push(d);
+                continue;
+            }
+            text = text.split(needle).join(`](${replacement})`);
+            rewritten += 1;
+            changed = true;
+        }
+        if (changed) {
+            fs.writeFileSync(abs, text, 'utf8');
+            filesTouched += 1;
+        }
+    }
+    return { rewritten, files: filesTouched, unresolved };
+}
+
+export function main(argv: string[] = process.argv.slice(2)): number {
+    if (argv.includes('-h') || argv.includes('--help')) {
+        process.stdout.write(
+            'usage: measure_docs_dead_links [--format json] [--list shipped|internal|py]\n' +
+                '                               [--fix shipped|py|all]\n' +
+                '  --fix rewrites a dead target only when EXACTLY ONE file in the tree\n' +
+                '        matches its path tail; ambiguity is reported, never guessed.\n',
+        );
+        return 0;
+    }
+    const asJson = argv.includes('--format') && argv[argv.indexOf('--format') + 1] === 'json';
+    const listIdx = argv.indexOf('--list');
+    const list = listIdx !== -1 ? argv[listIdx + 1] : null;
+    const fixIdx = argv.indexOf('--fix');
+    const fixScope = fixIdx !== -1 ? (argv[fixIdx + 1] ?? 'shipped') : null;
+
+    const r = measure();
+
+    if (fixScope !== null) {
+        const pool =
+            fixScope === 'shipped'
+                ? r.dead.filter((d) => d.cls === 'shipped')
+                : fixScope === 'py'
+                  ? r.dead.filter((d) => d.pyTarget)
+                  : r.dead;
+        const res = applyFix(REPO_ROOT, pool);
+        process.stdout.write(
+            `measure_docs_dead_links --fix ${fixScope}: rewrote ${String(res.rewritten)} link(s) in ` +
+                `${String(res.files)} file(s); ${String(res.unresolved.length)} left unresolved ` +
+                '(ambiguous or no unique candidate — never guessed).\n',
+        );
+        for (const d of res.unresolved.slice(0, 30)) {
+            process.stdout.write(`    unresolved: ${d.file} -> ${d.target}\n`);
+        }
+        return 0;
+    }
+    const shipped = r.dead.filter((d) => d.cls === 'shipped');
+    const internal = r.dead.filter((d) => d.cls === 'internal');
+    const py = r.dead.filter((d) => d.pyTarget);
+
+    if (list !== null && list !== undefined) {
+        const pick = list === 'shipped' ? shipped : list === 'internal' ? internal : py;
+        for (const d of pick) process.stdout.write(`${d.file} -> ${d.target}\n`);
+        return 0;
+    }
+    if (asJson) {
+        process.stdout.write(
+            `${JSON.stringify({ filesScanned: r.filesScanned, linksScanned: r.linksScanned, total: r.dead.length, shipped: shipped.length, internal: internal.length, pyTarget: py.length }, null, 2)}\n`,
+        );
+        return 0;
+    }
+    process.stdout.write(
+        `docs/ dead relative links — ${r.dead.length} of ${r.linksScanned} link(s) across ${r.filesScanned} file(s)\n` +
+            `  shipped   ${String(shipped.length).padStart(4)}  (package.json files[] publishes docs/guidelines/ plus three named contracts)\n` +
+            `  internal  ${String(internal.length).padStart(4)}\n` +
+            `  .py target${String(py.length).padStart(5)}  (migration leftovers; a subset of the two rows above, not a third bucket)\n`,
+    );
+    process.stdout.write(`scanned: ${r.filesScanned}\n`);
+    return 0;
+}
+
+/* c8 ignore start */
+if (process.argv[1] !== undefined && pathToFileURL(path.resolve(process.argv[1])).href === pathToFileURL(_HERE).href) {
+    process.exit(main());
+}
+/* c8 ignore stop */