@blamejs/exceptd-skills 0.19.33 → 0.19.34

package/scripts/bootstrap.js

@@ -2,45 +2,17 @@
 'use strict';
 
 /**
- * exceptd Security — bootstrap ceremony.
+ * Bootstrap ceremony. The mode is auto-detected, so a downstream consumer
+ * running it cannot invalidate the maintainer's signing key:
  *
- * Three audiences, three behaviors. The script auto-detects which mode is
- * appropriate so a downstream consumer never accidentally invalidates the
- * maintainer's signing key.
+ *   public key only     verify only — generates nothing, rewrites nothing
+ *   private key present re-sign every skill, then verify
+ *   no keys, or --init  generate an Ed25519 keypair, sign, verify
  *
- *   1. Downstream consumer (keys/public.pem exists, .keys/private.pem missing,
- *      no --init flag) — VERIFY ONLY. The maintainer already shipped the public
- *      key + signed manifest. Running `npm run bootstrap` here just confirms
- *      the working tree is intact. No keypair is generated; no signatures are
- *      rewritten. This is the safe default.
+ * The private key never leaves the maintainer's machine. keys/public.pem is the
+ * one tracked artifact, committed after first init.
  *
- *   2. Maintainer re-sign (.keys/private.pem exists) — SIGN + VERIFY. Used
- *      after editing skill content. Re-signs every skill with the existing
- *      private key, then verifies.
- *
- *   3. First-maintainer init (no keys/public.pem, OR --init explicitly passed)
- *      — GENERATE + SIGN + VERIFY. Used once when a maintainer sets up signing
- *      for a brand-new clone. Generates an Ed25519 keypair, signs every skill,
- *      and verifies. The new public key is committed; the private key stays in
- *      .keys/ (gitignored).
- *
- * The private key never leaves the maintainer's machine. The public key in
- * keys/public.pem is the one tracked artifact and is committed by the
- * maintainer after first init.
- *
- * Subprocesses use execFileSync (no shell) with argument arrays — there is no
- * user input on the path, and avoiding the shell removes the injection surface
- * regardless.
- *
- * Usage:
- *   node scripts/bootstrap.js              Auto-detect mode and run.
- *   node scripts/bootstrap.js --init       Force first-maintainer init (generate
- *                                          keypair + sign + verify).
- *   node scripts/bootstrap.js --force      Re-run even if marker exists.
- *   node scripts/bootstrap.js --help       Print this help text.
- *
- * Dependencies: Node 24 stdlib only. package.json has no runtime deps and
- * this script keeps it that way.
+ * Node stdlib only: package.json has no runtime deps and this keeps it so.
  */
 
 const fs = require('node:fs');
@@ -113,8 +85,8 @@ function run(label, scriptPath, scriptArgs) {
   const pretty = `node ${path.relative(ROOT, scriptPath)} ${scriptArgs.join(' ')}`.trim();
   console.log(`[bootstrap] ${label}: ${pretty}`);
   try {
-    // execFileSync: no shell, args passed as a vetted array. The script path
-    // and all args here are constants under this repo's control.
+    // execFileSync, never a shell: the path and args are repo constants passed
+    // as an array, so no injection surface exists.
     childProcess.execFileSync(process.execPath, [scriptPath, ...scriptArgs], {
       cwd: ROOT,
       stdio: 'inherit'
@@ -153,10 +125,8 @@ function main() {
   const mode = detectMode(args);
 
   if (mode === 'verify-only') {
-    // Downstream consumer path. The maintainer already shipped the public
-    // key and signed manifest. Running bootstrap here just confirms tree
-    // integrity — never generates or signs, which would invalidate the
-    // upstream maintainer's signing chain.
+    // Confirms tree integrity and nothing more. Generating or signing here
+    // would invalidate the upstream maintainer's signing chain.
     console.log('[bootstrap] Detected downstream-consumer state:');
     console.log('  - keys/public.pem present (shipped by maintainer)');
     console.log('  - .keys/private.pem absent');
@@ -170,8 +140,6 @@ function main() {
   }
 
   if (mode === 'resign') {
-    // Maintainer re-sign path. Private key already exists; re-sign every
-    // skill against the current content and verify.
     console.log('[bootstrap] Detected maintainer re-sign state (private key present).');
     console.log('[bootstrap] Re-signing every skill with the existing private key.');
     console.log();

package/scripts/build-indexes.js

@@ -1,56 +1,9 @@
 "use strict";
 /**
- * scripts/build-indexes.js
- *
- * Produces pre-computed indexes under `data/_indexes/` so AI consumers
- * and downstream tooling don't have to scan every skill + catalog
- * to answer routine cross-reference questions.
- *
- * Outputs (17 total):
- *   xref.json                — inverted index over cwe/d3fend/framework_gap/
- *                              atlas/attack/rfc/dlp citations
- *   trigger-table.json       — flat trigger → [skills]
- *   chains.json              — pre-computed cross-walks per CVE and per CWE
- *   jurisdiction-map.json    — jurisdiction → skills that reference it
- *   handoff-dag.json         — cross-skill mention graph
- *   summary-cards.json       — per-skill 100-word abstract
- *   section-offsets.json     — per-skill byte/line offsets of every H2
- *   token-budget.json        — approximate token cost per skill + section
- *   recipes.json             — curated multi-skill recipes
- *   jurisdiction-clocks.json — normalized obligation × hours matrix
- *   did-ladders.json         — canonical defense-in-depth ladders
- *   theater-fingerprints.json — structured compliance-theater pattern records
- *   currency.json            — pre-computed skill currency snapshot
- *   frequency.json           — citation-count tables per catalog field
- *   activity-feed.json       — "what changed when" feed
- *   catalog-summaries.json   — compact per-catalog summary cards
- *   stale-content.json       — persisted stale-content findings
- *   _meta.json               — SHA-256 of every source file for staleness
- *
- * Flags:
- *   (default)              build all outputs
- *   --only <names>         build only the comma-separated outputs (and
- *                          anything they depend on)
- *   --changed              build only outputs whose declared deps changed
- *                          since the last _meta.json snapshot. Safe in CI:
- *                          identical inputs always produce identical outputs.
- *   --parallel             run independent builders concurrently via
- *                          Promise.all (I/O concurrency, no worker threads).
- *                          For CPU-bound fan-out, callers can compose with
- *                          lib/worker-pool.js directly.
- *   --quiet                suppress per-output log lines
- *
- * Re-build conditions:
- *   _meta.json records sha256 of every source file. validate-indexes
- *   (predeploy gate) re-hashes those and fails if any source changed
- *   after the last build. --changed reads that same table to decide what
- *   to rebuild.
- *
- * Index file naming convention: leading underscore marks them as derived
- * (mirroring `_meta` in catalog files), so anyone scanning `data/` for
- * primary data filters them out.
- *
- * Node 24 stdlib only — zero npm deps.
+ * Produces the pre-computed indexes under `data/_indexes/`. Identical inputs must
+ * always produce identical outputs: --changed and the validate-indexes predeploy
+ * gate both key on the sha256 table in _meta.json, so a no-op run must not
+ * re-stamp.
  */
 
 const fs = require("fs");
@@ -69,20 +22,13 @@ function sha256(buf) {
 }
 
 function writeJson(name, obj) {
-  // Atomic write: a crash / disk-full / SIGKILL mid-write would otherwise leave
-  // a truncated JSON output on disk. Write to a temp sibling and rename — rename
-  // is atomic on POSIX and Windows, so a reader (or the next build's
-  // parse-check) only ever sees the complete old or complete new file.
+  // Temp sibling then rename, so a crash mid-write cannot leave a truncated index.
   const abs = path.join(IDX, name);
-  // Random suffix (not just pid) so a Promise.all fan-out (--parallel) sharing
-  // a pid can't race the same tmp path, and the in-repo temp name isn't
-  // predictable. Mirrors lib/citation-resolve.js cachePut.
+  // The random suffix keeps a --parallel fan-out sharing one pid off this path.
   const tmp = `${abs}.tmp-${process.pid}.${crypto.randomBytes(4).toString("hex")}`;
   fs.writeFileSync(tmp, JSON.stringify(obj, null, 2) + "\n", "utf8");
-  // rename-over-existing is atomic, but on Windows a sync client / AV / indexer
-  // (e.g. Dropbox holding the target open for a few ms) can make it transiently
-  // EPERM/EACCES/EBUSY. Retry with short backoff so a build doesn't fail on a
-  // lock that clears in milliseconds; on POSIX the first attempt always wins.
+  // On Windows a sync client or indexer holding the target open makes the rename
+  // transiently EPERM/EACCES/EBUSY, so it is retried with backoff.
   let lastErr;
   for (let attempt = 0; attempt < 10; attempt++) {
     try { fs.renameSync(tmp, abs); return; }
@@ -92,7 +38,6 @@ function writeJson(name, obj) {
       Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, 20 * (attempt + 1));
     }
   }
-  // Persistent failure: drop the temp sibling so it doesn't orphan, then surface.
   try { fs.unlinkSync(tmp); } catch { /* best effort */ }
   throw lastErr;
 }
@@ -133,15 +78,9 @@ Examples:
 `);
 }
 
-// --- Source loading (shared in-memory snapshot) -------------------------
-
-// Cross-reference fields the derived indexes key on. The manifest carries a
-// cache of these, but the skill frontmatter is the authoritative source — the
-// linter and staleness gate read frontmatter. Overlaying the parsed
-// frontmatter onto each skill record here means the indexes reflect the skill
-// bodies even when the manifest cache has drifted (e.g. dropping UK-CAF / AU
-// control mappings from framework_gaps). Array fields are overlaid only when
-// present in frontmatter; description is a scalar.
+// Skill frontmatter, not the manifest cache, is authoritative for these fields,
+// so the indexes stay aligned with the skill bodies when the cache drifts. A
+// field is overlaid only when frontmatter carries it.
 const FRONTMATTER_ARRAY_FIELDS = [
   "framework_gaps", "d3fend_refs", "cwe_refs", "atlas_refs",
   "attack_refs", "rfc_refs", "triggers", "data_deps",
@@ -169,11 +108,8 @@ function loadSources() {
   const skillBodies = {};
   for (const s of manifest.skills) skillBodies[s.name] = fs.readFileSync(ABS(s.path), "utf8");
 
-  // Build the skill records from the authoritative frontmatter, falling back
-  // to the manifest cache for fields frontmatter doesn't carry (signatures,
-  // dlp_refs, etc.). Downstream builders read cross-reference arrays from
-  // these records, so this is the single point that keeps the indexes aligned
-  // with the skill bodies.
+  // Frontmatter first, manifest cache for the fields frontmatter doesn't carry
+  // (signatures, dlp_refs). Every downstream builder reads these records.
   const skills = manifest.skills.map((s) => authoritativeSkill(s, skillBodies[s.name]));
   const skillNames = new Set(skills.map((s) => s.name));
 
@@ -196,15 +132,10 @@ function loadSources() {
   return ctx;
 }
 
-// --- Outputs registry ---------------------------------------------------
-// Each entry: { name, file, deps, build, dependsOn?: [name, ...] }
-//   deps: list of source-file pattern functions. A pattern is a function that
-//         returns true if a given relative path counts as a dep for this
-//         output. The --changed planner walks every changed source and
-//         flags every output whose deps match.
-//   dependsOn: names of other outputs that must be built first (used by
-//              chains.json which composes CVE + CWE halves, and
-//              token-budget.json which consumes section-offsets).
+// Registry entry: { name, file, deps, build, dependsOn?: [name, ...] }
+//   deps      predicates over a relative source path; --changed rebuilds an
+//             output when any changed source matches one.
+//   dependsOn output names built first, because this builder reads their file.
 
 function isAnySkillBody(p) { return p.startsWith("skills/") && p.endsWith("/skill.md"); }
 function isManifest(p) { return p === "manifest.json"; }
@@ -291,12 +222,8 @@ const OUTPUTS = [
       for (const s of ctx.skills) {
         const body = ctx.skillBodies[s.name];
         for (const code of codes) {
-          // Skip bare 2-letter ISO codes in free-text matching. `\bID\b`,
-          // `\bCA\b`, `\bNO\b` etc. collide with prose words ("the ID",
-          // "US-based") and control/countermeasure id grammar (`\bCA\b` matches
-          // inside `D3-CA`, `\bSA\b` inside `SA-12`), polluting coverage
-          // (Indonesia landed on 41/51 skills). These jurisdictions are mapped
-          // via the curated NAME_TO_CODE regulation-name table below instead.
+          // Bare 2-letter ISO codes never free-text match: `\bCA\b` collides with
+          // prose and with control-id grammar. They reach the map via NAME_TO_CODE.
           if (code.length <= 2) continue;
           const re = new RegExp("\\b" + code + "\\b");
           if (re.test(body) && !out[code].skills.includes(s.name)) out[code].skills.push(s.name);
@@ -361,9 +288,8 @@ const OUTPUTS = [
           entry_types: ["CVE", "CWE"],
         },
       };
-      // CVE half. Skip draft (_draft) CVEs so this derivation agrees with the
-      // reverse-ref index (refresh-reverse-refs.js excludes drafts) — both must
-      // reflect the same operator-queryable, curated truth.
+      // Drafts are skipped so this agrees with the reverse-ref index
+      // (scripts/refresh-reverse-refs.js), which excludes them too.
       for (const cveId of Object.keys(ctx.cveCatalog).filter((k) => !k.startsWith("_") && ctx.cveCatalog[k] && ctx.cveCatalog[k]._draft !== true)) {
         const cve = ctx.cveCatalog[cveId];
         const referencingSkills = ctx.skills
@@ -399,7 +325,6 @@ const OUTPUTS = [
           referencing_skills: referencingSkills, chain: hydrated,
         };
       }
-      // CWE half (delegated to builder)
       const cweChains = buildCweChains({
         skills: ctx.skills, cweCatalog: ctx.cweCatalog, atlasTtps: ctx.atlasTtps,
         cveCatalog: ctx.cveCatalog, frameworkGaps: ctx.frameworkGaps,
@@ -437,8 +362,7 @@ const OUTPUTS = [
     dependsOn: ["section-offsets"],   // needs the produced index
     build: (ctx) => {
       const { buildTokenBudget } = require("./builders/token-budget");
-      // section-offsets output is already on disk (built first by the
-      // dependency planner). Read it back for the token splitter.
+      // Already on disk: the dependency planner builds section-offsets first.
       const sectionOffsets = readJson(path.join(IDX, "section-offsets.json"));
       return buildTokenBudget({ root: ctx.root, skills: ctx.skills, sectionOffsets });
     },
@@ -541,8 +465,6 @@ const OUTPUTS = [
   },
 ];
 
-// --- Plan + run --------------------------------------------------------
-
 function loadPriorMeta() {
   const p = path.join(IDX, "_meta.json");
   if (!fs.existsSync(p)) return null;
@@ -552,9 +474,7 @@ function loadPriorMeta() {
 function liveSourceSet(ctx) {
   const out = new Set();
   out.add("manifest.json");
-  // README.md is consumed by the stale-content builder (badge-count drift), so
-  // it must be a hashed source — otherwise a README edit is invisible to
-  // --changed and the validate-indexes freshness gate.
+  // README.md feeds the stale-content builder, so it has to be a hashed source.
   if (fs.existsSync(ABS("README.md"))) out.add("README.md");
   for (const c of ctx.catalogFiles) out.add(c);
   for (const s of ctx.skills) out.add(s.path);
@@ -562,9 +482,7 @@ function liveSourceSet(ctx) {
 }
 
 function changedSources(ctx, priorMeta) {
-  // Returns the array of source paths whose sha256 differs from prior, OR
-  // every source if there's no prior meta. Also accounts for new + removed
-  // source files (which always force a rebuild).
+  // Every source when there is no prior meta; an appeared or disappeared one counts.
   if (!priorMeta || !priorMeta.source_hashes) {
     return [...liveSourceSet(ctx)];
   }
@@ -575,7 +493,6 @@ function changedSources(ctx, priorMeta) {
     const h = sha256(fs.readFileSync(ABS(p)));
     if (priorMeta.source_hashes[p] !== h) changed.push(p);
   }
-  // Any source that disappeared since last build counts as a change.
   for (const p of recorded) if (!live.has(p)) changed.push(p);
   return changed;
 }
@@ -590,12 +507,8 @@ function outputsAffectedBy(changedPaths) {
   return affected;
 }
 
-// An output's absence/corruption is itself a rebuild trigger. --changed keys
-// on source-hash deltas, but a derived file that was deleted, truncated, or
-// left unparseable must be regenerated even when every source is byte-identical
-// — otherwise the no-op path leaves a broken tree while writeMeta records it as
-// fresh (and reports a 0 count for the file it never rebuilt). Mirrors the
-// existence/parse check in lib/validate-indexes.js verifyOutputs.
+// A deleted, truncated or unparseable output is a rebuild trigger on its own:
+// --changed sees only source-hash deltas, so writeMeta would record it as fresh.
 function outputsMissingOrCorrupt() {
   const broken = new Set();
   for (const o of OUTPUTS) {
@@ -610,7 +523,6 @@ function outputsMissingOrCorrupt() {
 }
 
 function withDependencyClosure(names) {
-  // Pull in any dependsOn entries (e.g. token-budget needs section-offsets).
   const closure = new Set(names);
   let added = true;
   while (added) {
@@ -644,12 +556,10 @@ function topoOrder(names) {
 }
 
 async function runBuilders(ctx, names, opts) {
-  // Build the dependency-respecting execution plan, then dispatch.
   const order = topoOrder(names);
   const log = (s) => opts.quiet || console.log(s);
 
-  // Group by levels — outputs with no produced-output deps go first, then
-  // outputs depending on those, etc. This is the parallelization unit.
+  // A level holds outputs with no unbuilt dependsOn; --parallel fans out over one.
   const remaining = new Set(order);
   const levels = [];
   while (remaining.size > 0) {
@@ -693,13 +603,9 @@ function writeMeta(ctx, results, opts = {}) {
   const computedSourceHashes = {};
   for (const p of sourceFiles) computedSourceHashes[p] = sha256(fs.readFileSync(ABS(p)));
 
-  // A `--only` build regenerates an ARBITRARY subset of outputs (not the set
-  // implied by source changes), so the un-rebuilt outputs are not guaranteed to
-  // be consistent with the current sources. Preserve the prior source_hashes in
-  // that case, so validate-indexes still detects drift and demands a full
-  // rebuild — overwriting them with current hashes would fail OPEN, recording
-  // stale outputs as fresh. (--changed rebuilds exactly the change-affected
-  // outputs and a full build rebuilds all, so their current hashes are accurate.)
+  // A `--only` build touches an arbitrary subset, so keeping the prior hashes
+  // lets validate-indexes still demand a full rebuild. Writing current hashes
+  // here fails open, recording stale outputs as fresh.
   const partial = !!(opts && opts.only);
   const sourceHashes = (partial && prior && prior.source_hashes && typeof prior.source_hashes === "object")
     ? prior.source_hashes
@@ -707,15 +613,8 @@ function writeMeta(ctx, results, opts = {}) {
 
   const outputsList = OUTPUTS.map((o) => o.file).sort();
 
-  // Determinism: preserve the prior `generated_at` when the inputs are
-  // byte-identical to the last build (same source hashes AND same output set).
-  // validate-indexes keys freshness on `source_hashes`/`outputs`, never on the
-  // timestamp, so re-stamping a no-op run with a fresh wall-clock value adds
-  // no freshness signal and only produces a spurious git diff — contradicting
-  // the documented "identical inputs always produce identical outputs"
-  // contract for --changed (and the idempotence the predeploy gate relies on).
-  // Reuse the prior timestamp on a genuine no-op; mint a new one only when the
-  // hashed surface actually moved (or there is no prior meta to inherit from).
+  // `generated_at` carries over when the source hashes and the output set are
+  // identical to the last build, so a no-op run produces no git diff.
   const sourcesUnchanged = prior && prior.source_hashes &&
     JSON.stringify(prior.source_hashes) === JSON.stringify(sourceHashes) &&
     JSON.stringify(prior.outputs || []) === JSON.stringify(outputsList);
@@ -723,8 +622,7 @@ function writeMeta(ctx, results, opts = {}) {
     ? prior.generated_at
     : new Date().toISOString();
 
-  // Stats are computed from in-memory results when available, else from disk
-  // (covers --only / --changed runs that didn't rebuild every output).
+  // In-memory result when this run built it, otherwise from disk.
   function readBack(name) {
     if (results[name]) return results[name];
     const o = OUTPUTS.find((x) => x.name === name);
@@ -764,20 +662,16 @@ function writeMeta(ctx, results, opts = {}) {
     source_hashes: sourceHashes,
     skill_count: ctx.skills.length,
     catalog_count: ctx.catalogFiles.length,
-    // The derived index files this build produces. validate-indexes confirms
-    // every one still exists and parses — source hashes alone do not detect a
-    // deleted/truncated/corrupted OUTPUT (an index file removed from a clean
-    // source tree would otherwise pass the freshness gate as "current").
+    // Source hashes alone miss a deleted output: removed from a clean source
+    // tree, it would pass the freshness gate as current.
     outputs: outputsList,
     index_stats: {
       xref_entries: xrefStats,
       trigger_table_entries: Object.keys(trigger).length,
       chains_cve_entries: cveChainCount,
       chains_cwe_entries: cweChainCount,
-      // The jurisdiction-map indexes EVERY jurisdiction code (including those
-      // with no breach/patch clocks), so it — not the clocks table — is the
-      // count of jurisdictions actually indexed. (`jurisdiction_clocks` below
-      // reports the clocks-only subset.)
+      // Counted off jurisdiction-map, which holds every code; `jurisdiction_clocks`
+      // below is the clocks-only subset.
       jurisdictions_indexed: Object.keys(readBack("jurisdiction-map") || {}).length || Object.keys(jurisdictionClocks.by_jurisdiction || {}).length,
       handoff_dag_nodes: handoff.nodes?.length || 0,
       summary_cards: Object.keys(summaryCards.skills || {}).length,
@@ -806,7 +700,6 @@ async function main() {
   const ctx = loadSources();
   const log = (s) => opts.quiet || console.log(s);
 
-  // Decide which outputs to build.
   let chosen;
   if (opts.only) {
     const wanted = opts.only.split(",").map((s) => s.trim()).filter(Boolean);
@@ -822,22 +715,15 @@ async function main() {
     const changed = changedSources(ctx, prior);
     log(`changed sources: ${changed.length}`);
     const affected = outputsAffectedBy(changed);
-    // Union in any output that no longer exists on disk or fails to parse.
-    // A deleted/corrupt output is a rebuild trigger independent of source
-    // hashes — without this, --changed reports "nothing to do" and writeMeta
-    // would claim freshness for a tree that's actually broken.
+    // Without this union, --changed reports "nothing to do" on a corrupt output.
     const broken = outputsMissingOrCorrupt();
     for (const name of broken) affected.add(name);
     if (broken.size > 0) log(`missing/corrupt outputs to regenerate: ${[...broken].sort().join(", ")}`);
     chosen = withDependencyClosure(affected);
     if (chosen.size === 0) {
       log("build-indexes: no outputs need rebuilding (sources unchanged)");
-      // Rewrite _meta.json so its hash table + outputs list self-heal against
-      // the live source set (e.g. an old-format _meta that predates a field).
-      // writeMeta preserves the prior `generated_at` when the hashed surface is
-      // byte-identical, so a genuine no-op leaves the file unchanged — no
-      // spurious timestamp-only diff. Re-stamping would add no freshness signal
-      // (validate-indexes keys on source_hashes/outputs, not the timestamp).
+      // Still rewritten, so the hash table and outputs list self-heal against the
+      // live source set; a true no-op leaves the file byte-identical.
       writeMeta(ctx, {}, opts);
       return;
     }

package/scripts/builders/activity-feed.js

@@ -1,17 +1,8 @@
 "use strict";
 /**
- * scripts/builders/activity-feed.js
- *
- * Builds `data/_indexes/activity-feed.json` — a "what changed when" feed
- * across skills and catalogs, sorted by date. Lightweight RSS for the
- * skill corpus that consumers can poll without diff-ing the manifest.
- *
- * Combines:
- *   - per-skill last_threat_review
- *   - per-catalog _meta.last_updated
- *   - manifest threat_review_date + atlas_version_date when present
- *
- * Output sorted descending by date.
+ * Builds `data/_indexes/activity-feed.json`: a "what changed when" feed across
+ * skills and catalogs, sorted descending by date, so a consumer can poll it
+ * rather than diff the manifest.
  */
 
 const fs = require("fs");
@@ -49,8 +40,7 @@ function buildActivityFeed({ root, manifest, skills, catalogFiles }) {
         });
       }
     } catch {
-      // skip non-JSON or malformed; build-indexes runs after lint so this
-      // is unlikely.
+      // Skip malformed JSON; lint runs before build-indexes.
     }
   }
 

package/scripts/builders/catalog-summaries.js

@@ -1,15 +1,8 @@
 "use strict";
 /**
- * scripts/builders/catalog-summaries.js
- *
- * Builds `data/_indexes/catalog-summaries.json` — for each data/<catalog>.json
- * file, a compact summary: purpose, entry count, version pin (where applicable),
- * source confidence, TLP, last-updated date. Consumers can load this single
- * file (~3-4 KB) instead of every _meta block to learn what catalogs are
- * available and how fresh they are.
- *
- * Curated human-readable purpose strings: keep these in lockstep with the
- * canonical catalog README in `docs/data-catalogs.md` if/when added.
+ * Builds `data/_indexes/catalog-summaries.json`: one compact card per
+ * data/<catalog>.json, so a consumer learns what catalogs exist and how fresh
+ * they are from a single small file rather than every _meta block.
  */
 
 const fs = require("fs");

package/scripts/builders/currency.js

@@ -1,29 +1,18 @@
 "use strict";
 /**
- * scripts/builders/currency.js
- *
- * Builds `data/_indexes/currency.json` — pre-computed currency scores for
- * every skill against a deterministic reference date (manifest's
- * `threat_review_date`). Saves the watchlist/scheduler from re-running
- * `orchestrator currency` to produce the same answer.
- *
- * The reference date is deterministic so the file hash stays stable until
- * skills or the reference change — which is what `validate-indexes`
- * requires. The orchestrator's interactive `currency` command remains the
- * source-of-truth for live decay; this index is the snapshot view.
- *
- * Decay formula matches pipeline.js _currencyScore() exactly:
- *   >180 days → -30, >90 → -20, >60 → -10, >30 → -5
- *   -5 per forward_watch entry
+ * Builds `data/_indexes/currency.json` — per-skill currency scored against
+ * manifest.threat_review_date rather than today, so the file hash stays
+ * stable until skills or the reference change, which is what
+ * `validate-indexes` requires. The orchestrator's `currency` command is the
+ * live-decay view; this index is the snapshot.
  */
 
 const fs = require("fs");
 const path = require("path");
 
 function currencyScore(daysSinceReview, _forwardWatchCount) {
-  // See orchestrator/pipeline.js — forward_watch count no longer
-  // affects currency score (it's a maintenance signal, not staleness).
-  // Param retained for ABI compatibility with callers.
+  // Scoring matches orchestrator/pipeline.js _currencyScore(). forward_watch
+  // count is a maintenance signal, not staleness, so the param is unused.
   let score = 100;
   if (daysSinceReview > 180) score -= 30;
   else if (daysSinceReview > 90) score -= 20;
@@ -43,8 +32,6 @@ function parseFrontmatterForwardWatchCount(body) {
   const m = body.match(/^---\n([\s\S]*?)\n---/);
   if (!m) return 0;
   const fm = m[1];
-  // Counts top-level "forward_watch:" list items. Lines starting with "  - "
-  // immediately after a "forward_watch:" line until the next non-indented key.
   const lines = fm.split(/\r?\n/);
   let inFw = false;
   let count = 0;

package/scripts/builders/cwe-chains.js

@@ -1,29 +1,11 @@
 "use strict";
 /**
- * scripts/builders/cwe-chains.js
+ * Builds the CWE-keyed half of `data/_indexes/chains.json`, in the same
+ * hydrated cross-walk shape as the CVE-keyed half.
  *
- * Builds the CWE side of `data/_indexes/chains.json`. The existing chains
- * object is keyed by CVE-id; this builder produces CWE-id entries with the
- * same hydrated cross-walk shape so AI consumers can start from a CWE and
- * see every skill / catalog dimension that touches the weakness class.
- *
- * Per-CWE shape:
- *   {
- *     name:                CWE catalog entry name
- *     category:            CWE catalog category (memory-safety / injection / etc.)
- *     referencing_skills:  every skill listing this CWE in cwe_refs
- *     chain: {
- *       atlas, attack_refs, framework_gaps, d3fend, rfc_refs, dlp_refs
- *     }
- *     related_cves:        CVEs in cve-catalog.json whose framework gaps
- *                          surface via skills that also cite this CWE
- *   }
- *
- * The CWE → CVE link is indirect — CWEs aren't currently stamped on each
- * CVE entry directly. The skill bodies are the connective tissue. So we
- * compute it the same way the CVE chains builder does: skills cite CWEs,
- * skills cite framework_gaps, framework_gaps surface evidence_cves. The
- * intersection through the skill graph gives the related-CVE set.
+ * The CWE → CVE link is indirect: CWE ids are not stamped on CVE entries, so
+ * related_cves is computed through the skill graph — skills cite CWEs, skills
+ * cite framework_gaps, framework_gaps surface evidence_cves.
  */
 
 function buildCweChains({ skills, cweCatalog, atlasTtps, cveCatalog, frameworkGaps, d3fendCatalog, rfcCatalog }) {
@@ -37,7 +19,6 @@ function buildCweChains({ skills, cweCatalog, atlasTtps, cveCatalog, frameworkGa
       .filter((s) => (s.cwe_refs || []).includes(cweId))
       .map((s) => s.name);
 
-    // Aggregate cross-refs from those skills.
     const accum = {
       atlas_refs: new Set(),
       attack_refs: new Set(),
@@ -54,7 +35,6 @@ function buildCweChains({ skills, cweCatalog, atlasTtps, cveCatalog, frameworkGa
       }
     }
 
-    // Hydrate the cross-walk dimensions for the AI consumer.
     const hydrated = {
       atlas: [...accum.atlas_refs].sort().map((a) => ({
         id: a,
@@ -80,11 +60,8 @@ function buildCweChains({ skills, cweCatalog, atlasTtps, cveCatalog, frameworkGa
       dlp_refs: [...accum.dlp_refs].sort(),
     };
 
-    // Related CVEs: walk evidence_cves on the framework_gaps that the
-    // referencing skills cite. Inner join via the skill graph. Skip draft
-    // (_draft) CVEs so this CWE half agrees with the CVE half (build-indexes.js)
-    // and the reverse-ref index — all three must reflect the same curated,
-    // operator-queryable truth.
+    // Draft (_draft) CVEs are skipped so this half agrees with the CVE half
+    // (build-indexes.js) and the reverse-ref index.
     const relatedCves = new Set();
     for (const gap of accum.framework_gaps) {
       for (const ev of (frameworkGaps[gap]?.evidence_cves || [])) {