@adia-ai/adia-ui-forge 0.8.8 → 0.8.10

package/skills/adia-llm-internals/evals/routing-corpus.json

@@ -1,37 +1,134 @@
 {
-  "positives": [
-    "the gemini adapter is dropping usage tokens on streamed responses",
-    "stopReason from openai never reaches the client as MAX_TOKENS",
-    "the SSE parser chokes when a chunk splits mid-line",
-    "the passthrough proxy is returning 401 for anthropic requests in production",
-    "add a new model to the MODELS registry and set it as DEFAULT_MODEL",
-    "createAdapter() is leaking a real API key to the browser on a production host",
-    "add support for a new OpenAI-compatible gateway provider adapter",
-    "streamChat() never emits a terminal done chunk",
-    "detectProvider isn't resolving the new model id to the right adapter",
-    "the anthropic adapter's buildRequest() diverges from the passthrough proxy path",
-    "the StreamChunk union is missing a variant for tool_use deltas",
-    "the stub LLM adapter isn't returning parseable A2UI JSON in dev mode"
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "name": "adia-llm-internals routing accuracy corpus",
+  "version": "2.1.0",
+  "purpose": "Routing-eval corpus for adia-llm-internals. Each phrase declares the skill (expected), a forbidden skill (expected_not, for phrases the source data only ever asserted as \"not this skill\"), or neither. Scored by scripts/skills/run-skill-evals.mjs (TF-IDF token overlap over per-skill description+triggers).",
+  "scoring_notes": "Heuristic signal, not ground truth. Treat misroutes as a prompt to tighten the skill description, never as a reason to keyword-stuff it. Real harness routing is LLM-driven.",
+  "scope": "adia-llm-internals routing \u2014 does this phrase activate adia-llm-internals?",
+  "phrases": [
+    {
+      "id": "adia-llm-internals-pos-01",
+      "phrase": "the gemini adapter is dropping usage tokens on streamed responses",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-02",
+      "phrase": "stopReason from openai never reaches the client as MAX_TOKENS",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-03",
+      "phrase": "the SSE parser chokes when a chunk splits mid-line",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-04",
+      "phrase": "the passthrough proxy is returning 401 for anthropic requests in production",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-05",
+      "phrase": "add a new model to the MODELS registry and set it as DEFAULT_MODEL",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-06",
+      "phrase": "createAdapter() is leaking a real API key to the browser on a production host",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-07",
+      "phrase": "add support for a new OpenAI-compatible gateway provider adapter",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-08",
+      "phrase": "streamChat() never emits a terminal done chunk",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-09",
+      "phrase": "detectProvider isn't resolving the new model id to the right adapter",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-10",
+      "phrase": "the anthropic adapter's buildRequest() diverges from the passthrough proxy path",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-11",
+      "phrase": "the StreamChunk union is missing a variant for tool_use deltas",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-pos-12",
+      "phrase": "the stub LLM adapter isn't returning parseable A2UI JSON in dev mode",
+      "expected": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-01",
+      "phrase": "wire the chat client into our admin app's chat-shell",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-02",
+      "phrase": "why did the composer emit composition-synthesized instead of composition-match",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-03",
+      "phrase": "cut a release and bump @adia-ai/llm's version with the rest of the lockstep packages",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-04",
+      "phrase": "add a new prop to the chat-input-ui primitive",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-05",
+      "phrase": "explain how Server-Sent Events work in general",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-06",
+      "phrase": "add a chat box to the admin app",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-07",
+      "phrase": "the chat-shell message list isn't rendering streamed replies",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-08",
+      "phrase": "tune STRONG_MATCH and lift the eval fails in the compose strategies",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-09",
+      "phrase": "deploy the ui-kit dist to exe.dev and restart the service",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-10",
+      "phrase": "run a dogfood sweep to find broken demos",
+      "expected_not": "adia-llm-internals"
+    },
+    {
+      "id": "adia-llm-internals-neg-11",
+      "phrase": "score the gen-ui gallery outputs against the exit gate",
+      "expected_not": "adia-llm-internals"
+    }
   ],
-  "negatives": [
-    "wire the chat client into our admin app's chat-shell",
-    "why did the composer emit composition-synthesized instead of composition-match",
-    "cut a release and bump @adia-ai/llm's version with the rest of the lockstep packages",
-    "add a new prop to the chat-input-ui primitive",
-    "explain how Server-Sent Events work in general",
-    "add a chat box to the admin app",
-    "the chat-shell message list isn't rendering streamed replies",
-    "tune STRONG_MATCH and lift the eval fails in the compose strategies",
-    "deploy the ui-kit dist to exe.dev and restart the service",
-    "run a dogfood sweep to find broken demos",
-    "score the gen-ui gallery outputs against the exit gate"
-  ],
-  "_measured": {
+  "_measured_historical": {
     "as_of": "2026-07-18",
     "scorer": "routing_eval.py (nonoun-plugins/forge)",
     "f1": 0.857,
     "precision": 1.0,
     "recall": 0.75,
-    "exit_code": 0
+    "exit_code": 0,
+    "note": "measured by the external routing_eval.py (nonoun-plugins/forge) against the pre-conversion positives/negatives shape; historical record only, not regenerated by run-skill-evals.mjs."
   }
 }

package/skills/adia-release/evals/routing-corpus.json

@@ -1,35 +1,124 @@
 {
-  "positives": [
-    "Cut and ship v0.8.5 of the AdiaUI packages",
-    "The [Unreleased] CHANGELOG entries are piling up, let's cut a release",
-    "Tag the lockstep packages and push them to origin",
-    "check:lockstep is failing \u2014 one package didn't bump",
-    "Dispatch the publish workflows for v0.8.5 and verify npm",
-    "We have three unpushed release commits piled up, batch-push them",
-    "F-N1 is warning about a CHANGELOG entry not covering a touched directory",
-    "The release cut landed on the wrong branch, help me recover",
-    "Author the release notes and Slack post for the last two versions",
-    "This cut removes the variant prop from button-ui, write the MIGRATION GUIDE section",
-    "Publish the adia-ui-forge plugin independently of the lockstep set",
-    "Just verify the release gates are green without cutting anything",
-    "Tags are pushed but zero publish workflows fired, what happened"
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "name": "adia-release routing accuracy corpus",
+  "version": "2.1.0",
+  "purpose": "Routing-eval corpus for adia-release. Each phrase declares the skill (expected), a forbidden skill (expected_not, for phrases the source data only ever asserted as \"not this skill\"), or neither. Scored by scripts/skills/run-skill-evals.mjs (TF-IDF token overlap over per-skill description+triggers).",
+  "scoring_notes": "Heuristic signal, not ground truth. Treat misroutes as a prompt to tighten the skill description, never as a reason to keyword-stuff it. Real harness routing is LLM-driven.",
+  "scope": "adia-release routing \u2014 does this phrase activate adia-release?",
+  "phrases": [
+    {
+      "id": "adia-release-pos-01",
+      "phrase": "Cut and ship v0.8.5 of the AdiaUI packages",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-02",
+      "phrase": "The [Unreleased] CHANGELOG entries are piling up, let's cut a release",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-03",
+      "phrase": "Tag the lockstep packages and push them to origin",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-04",
+      "phrase": "check:lockstep is failing \u2014 one package didn't bump",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-05",
+      "phrase": "Dispatch the publish workflows for v0.8.5 and verify npm",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-06",
+      "phrase": "We have three unpushed release commits piled up, batch-push them",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-07",
+      "phrase": "F-N1 is warning about a CHANGELOG entry not covering a touched directory",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-08",
+      "phrase": "The release cut landed on the wrong branch, help me recover",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-09",
+      "phrase": "Author the release notes and Slack post for the last two versions",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-10",
+      "phrase": "This cut removes the variant prop from button-ui, write the MIGRATION GUIDE section",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-11",
+      "phrase": "Publish the adia-ui-forge plugin independently of the lockstep set",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-12",
+      "phrase": "Just verify the release gates are green without cutting anything",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-pos-13",
+      "phrase": "Tags are pushed but zero publish workflows fired, what happened",
+      "expected": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-01",
+      "phrase": "Restart the exe.dev VM, it's returning 502s",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-02",
+      "phrase": "Rsync the built dist folder to ui-kit.exe.xyz",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-03",
+      "phrase": "Sweep our consumer app for the renamed variant prop after the last breaking release",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-04",
+      "phrase": "Add a new size variant to the button-ui primitive",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-05",
+      "phrase": "Compose a settings screen for our app from the AdiaUI catalog",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-06",
+      "phrase": "zettel coverage dropped on the nightly eval, find out why",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-07",
+      "phrase": "sweep the component demos for native-primitive leaks before we cut",
+      "expected_not": "adia-release"
+    },
+    {
+      "id": "adia-release-neg-08",
+      "phrase": "the gemini adapter is dropping usage tokens on streamed responses",
+      "expected": "adia-llm-internals"
+    }
   ],
-  "negatives": [
-    "Restart the exe.dev VM, it's returning 502s",
-    "Rsync the built dist folder to ui-kit.exe.xyz",
-    "Sweep our consumer app for the renamed variant prop after the last breaking release",
-    "Add a new size variant to the button-ui primitive",
-    "Compose a settings screen for our app from the AdiaUI catalog",
-    "zettel coverage dropped on the nightly eval, find out why",
-    "sweep the component demos for native-primitive leaks before we cut",
-    "the gemini adapter is dropping usage tokens on streamed responses"
-  ],
-  "_measured": {
+  "_measured_historical": {
     "as_of": "2026-07-18",
     "scorer": "routing_eval.py (nonoun-plugins/forge)",
     "f1": 1.0,
     "precision": 1.0,
     "recall": 1.0,
-    "exit_code": 0
+    "exit_code": 0,
+    "note": "measured by the external routing_eval.py (nonoun-plugins/forge) against the pre-conversion positives/negatives shape; historical record only, not regenerated by run-skill-evals.mjs."
   }
 }

package/skills/adia-release/scripts/pr-bridge.mjs

@@ -124,6 +124,20 @@ export function bridgeDecision(state) {
     if (!PASS.has(concl)) reasons.push(`check '${c.name}' landed ${concl}`);
     else gatingLanded++;
   }
+  // gh#(pr-bridge draft guard): a draft PR is the AUTHOR's own "not ready"
+  // signal — GitHub's API refuses to merge one anyway (GraphQL: "Pull
+  // Request is still a draft"), and the old code had no isDraft check, so
+  // bridgeDecision said MERGE, the gh pr merge call threw, and the whole
+  // script crashed uncaught. A caller piping through `grep -E "pr-bridge" |
+  // tail -N` (this script's own documented usage pattern) sees only the
+  // last N lines BEFORE the crash and no error — reads as quiet success.
+  // Unconditional stop (review nit: placed after the per-check loop above,
+  // not literally first — but every reasons.push() here is unconditional and
+  // the "stop" decision below fires on ANY non-empty reasons list, so a
+  // draft overrides an otherwise-perfect merge regardless of ordering):
+  // never attempt the merge call on a draft, no matter how green everything
+  // else is.
+  if (state.isDraft) reasons.push('PR is a DRAFT — the author has not marked it ready for review; never auto-merge a draft');
   if (state.reviewDecision === 'CHANGES_REQUESTED') reasons.push('a review requests changes');
   if (state.unresolvedThreads > 0) reasons.push(`${state.unresolvedThreads} unresolved review thread(s) — admin merge cannot bypass these`);
   if (state.mergeable === 'CONFLICTING') reasons.push('PR is CONFLICTING with the base branch');
@@ -144,7 +158,7 @@ export function bridgeDecision(state) {
 // ── GitHub reads ──────────────────────────────────────────────────
 function readPrState(prNumber) {
   const view = JSON.parse(sh(
-    `gh pr view ${prNumber} --json state,mergeable,reviewDecision,statusCheckRollup,headRefName,baseRefName`,
+    `gh pr view ${prNumber} --json state,isDraft,mergeable,reviewDecision,statusCheckRollup,headRefName,baseRefName`,
   ));
   const checks = (view.statusCheckRollup || []).map((c) => ({
     name: c.name || c.context || '?',
@@ -156,6 +170,7 @@ function readPrState(prNumber) {
   )).data.repository.pullRequest.reviewThreads.nodes;
   return {
     state: view.state,
+    isDraft: view.isDraft,
     headRefName: view.headRefName,
     baseRefName: view.baseRefName,
     checks,
@@ -240,7 +255,19 @@ async function main() {
     }
     if (d.action === 'merge') {
       console.log(`\n[pr-bridge] all gates green — merging PR #${prNumber} (admin merge per house flow §5.7)`);
-      sh(`gh pr merge ${prNumber} --admin --merge --delete-branch`);
+      // Defense in depth: the isDraft check above should catch this, but
+      // GitHub's own merge call is the actual authority — never let its
+      // rejection crash uncaught (a caller filtering output for the
+      // "pr-bridge" prefix, this script's own documented usage, would see
+      // the run go quiet mid-merge with no clear failure, easy to misread
+      // as success). Report loudly and exit 1 instead of throwing.
+      try {
+        sh(`gh pr merge ${prNumber} --admin --merge --delete-branch`);
+      } catch (e) {
+        console.error(`\n[pr-bridge] STOP — gh pr merge rejected PR #${prNumber}: ${e.message || e}`);
+        console.error('  Nothing else mutated by this run; investigate and merge manually if appropriate.');
+        process.exit(1);
+      }
       break;
     }
     if (Date.now() > deadline) {
@@ -269,10 +296,12 @@ async function main() {
 // ── Selftest — the predicate is the load-bearing part ─────────────
 function selftest() {
   const ck = (name, status, conclusion) => ({ name, status, conclusion });
-  const base = { unresolvedThreads: 0, reviewDecision: 'REVIEW_REQUIRED', mergeable: 'MERGEABLE' };
+  const base = { unresolvedThreads: 0, reviewDecision: 'REVIEW_REQUIRED', mergeable: 'MERGEABLE', isDraft: false };
   const cases = [
     ['all green → merge',
       { ...base, checks: [ck('Static checks + MCP smoke', 'COMPLETED', 'SUCCESS'), ck('CodeRabbit', '', 'SUCCESS')] }, 'merge'],
+    ['DRAFT PR with all-green checks → stop, never attempt gh pr merge (real incident: PR #362)',
+      { ...base, isDraft: true, checks: [ck('Static checks + MCP smoke', 'COMPLETED', 'SUCCESS'), ck('CodeRabbit', '', 'SUCCESS')] }, 'stop'],
     ['fail-soft failure is advisory → merge',
       { ...base, checks: [ck('Static checks + MCP smoke', 'COMPLETED', 'SUCCESS'), ck('Audit health (fail-soft)', 'COMPLETED', 'FAILURE')] }, 'merge'],
     ['EMPTY check set → wait, never blind-merge (reviewer finding #1)',
@@ -313,4 +342,12 @@ function selftest() {
 
 const argv = process.argv.slice(2);
 if (argv[0] === 'selftest') selftest();
-else main().catch((e) => { console.error(e.message || e); process.exit(1); });
+else main().catch((e) => {
+  // Reviewer finding (PR #364): every OTHER error path (readPrState()'s gh
+  // pr view/graphql calls on every poll, git push/gh pr create, the
+  // re-baseline sequence) throws up to this top-level handler, which had NO
+  // "[pr-bridge]" prefix — invisible under the exact `grep -E "pr-bridge" |
+  // tail -N` filter this PR exists to fix. Same class, different call site.
+  console.error(`[pr-bridge] STOP — ${e.message || e}`);
+  process.exit(1);
+});

package/skills/adia-release/scripts/release-pack.mjs

@@ -229,6 +229,21 @@ function shCapture(cmd, args) {
 // ── Step 1 — Re-baseline ──────────────────────────────────────────
 function step1ReBaseline(args) {
   console.log('\n=== Step 1 — Re-baseline ===');
+  // cut-procedure.md §Step 1 requires `git branch --show-current` == main
+  // before a cut starts — cutting on any other branch pushes a stale main ref
+  // while tags point at the wrong tip (recovery-paths.md §Scenario 8). The doc
+  // asserted this; nothing here ever checked it (gh#358). The one legitimate
+  // exception is resuming this SAME cut after step5Commit already created and
+  // checked out release/vX.Y.Z — allow only that exact branch name back in.
+  if ((args.mode === 'cut' || args.mode === 'from-scratch') && !args.dry) {
+    const branch = shQuiet('git branch --show-current', args).trim();
+    const releaseBranch = `release/v${args.version}`;
+    if (branch !== 'main' && branch !== releaseBranch) {
+      console.error(`ERROR: on branch '${branch}' — cut modes start from 'main' (or resume on '${releaseBranch}'). Switch branches before proceeding.`);
+      process.exit(1);
+    }
+    console.log(`  branch: ${branch} (ok)`);
+  }
   console.log('git status --short:');
   sh('git status --short', args);
   console.log('\norigin/main..HEAD (unpushed):');
@@ -362,6 +377,30 @@ function step5Commit(args) {
     return;
   }
   console.log('\n=== Step 5 — Stage + Commit ===');
+  // cut-procedure.md §Step 5: "Branch FIRST, then stage" — this step never
+  // actually created release/vX.Y.Z, so the commit landed on whatever branch
+  // was checked out (main, on the v0.8.8 cut — gh#358). Idempotent: if the
+  // branch already exists (a resumed run), check it out rather than failing;
+  // step1ReBaseline's branch assertion is what stops a genuinely wrong branch
+  // from reaching this point.
+  const releaseBranch = `release/v${args.version}`;
+  const currentBranch = args.dry ? null : shQuiet('git branch --show-current', args).trim();
+  if (currentBranch !== releaseBranch) {
+    const exists = !args.dry && shQuiet(`git branch --list ${releaseBranch}`, args).trim().length > 0;
+    // Reviewer finding (PR #359): checking out an EXISTING release branch (a
+    // resumed run) on a dirty tree fails with a raw, opaque git error — Step 4
+    // (promote/bump/lockfile) runs immediately before this and dirties exactly
+    // the files a release branch differs from main in. A fresh `-b` create
+    // from clean main can't hit this, so the check only guards the resume path.
+    if (exists && !args.dry) {
+      const dirty = shQuiet('git status --porcelain', args).trim();
+      if (dirty) {
+        console.error(`ERROR: '${releaseBranch}' already exists, but the working tree is dirty — checking out would fail or silently carry these changes onto the resumed branch:\n${dirty}\n  Commit, stash, or discard first, then re-run.`);
+        process.exit(1);
+      }
+    }
+    sh(`git checkout ${exists ? '' : '-b '}${releaseBranch}`, args);
+  }
   sh('git reset HEAD >/dev/null 2>&1 || true', args);
   const releaseFiles = [
     'package-lock.json',
@@ -512,11 +551,34 @@ async function step9Publish(args) {
   // Give GH a moment to index the runs created by the per-tag push in Step 8.
   if (!args.dry) sh('sleep 6', args);
   sh(`node ${SCRIPT_DIR}/dispatch-publish.mjs --version ${args.version} --scope ${args.scope} --verify-triggered`, args);
-  console.log('\n  Waiting for publishes to settle...');
-  sh(`until [ "$(gh run list --workflow=publish-a2ui-validator.yml --limit 1 --json status -q '.[0].status')" = "completed" ]; do sleep 5; done`, args);
-  console.log('  ✓ publishes settled');
+  // gh#354: this used to poll ONE workflow (publish-a2ui-validator.yml) as a
+  // sample-of-one proxy for all 11 publishes settling — a straggler on any
+  // OTHER package's workflow made the fail-closed check below abort on a pure
+  // timing false alarm (hit live on the v0.8.8 cut: a2ui-retrieval published
+  // successfully ~1 minute after validator did, after the old proxy had
+  // already declared "settled" and the registry read caught it mid-flight).
+  // Poll the REGISTRY itself instead — the skill's own doctrine (SKILL.md:
+  // "the registry is the verify target, never a workflow's green check")
+  // applies here as much as it does to the final check.
+  console.log(`\n  Waiting for the registry to confirm all ${PACKAGES.length} packages at ${args.version} (polling every 15s, up to 10min)...`);
+  sh(
+    `for i in $(seq 1 40); do ` +
+    `fail=0; ` +
+    `for pkg in ${PACKAGES.join(' ')}; do ` +
+    `v="$(npm view "${args.scope}/$pkg" version 2>/dev/null || true)"; ` +
+    `[ "$v" = "${args.version}" ] || fail=1; ` +
+    `done; ` +
+    `latest="$(npm view ${args.scope}/web-components dist-tags.latest 2>/dev/null || true)"; ` +
+    `[ "$latest" = "${args.version}" ] || fail=1; ` +
+    `[ "$fail" = 0 ] && { echo "  ✓ registry confirms all ${PACKAGES.length} + latest, after $i check(s)"; exit 0; }; ` +
+    `sleep 15; ` +
+    `done; ` +
+    `echo "  registry did not converge after 10min — the fail-closed check below will report exactly which package(s) are still stale"`,
+    args,
+  );
   // Fail-closed: the registry is the verify target (SKILL.md) — a missing or
-  // stale package aborts the cut here, before GH releases + site deploy.
+  // stale package aborts the cut here, before GH releases + site deploy. This
+  // is now a final confirmation of the poll above, not the first real read.
   console.log(`\n  Post-publish registry verification, all ${PACKAGES.length} at ${args.version} + npm latest (fail-closed):`);
   sh(
     `fail=0; ` +
@@ -648,6 +710,8 @@ function selftest() {
       `${GATE_ROSTER.length}-gate roster`,
       `✓ all ${GATE_ROSTER.length}/${GATE_ROSTER.length} gates passed`,
       'STOPPING per invariant 3',
+      // gh#358 — the release branch must actually be created, not just documented.
+      'git checkout -b release/v9.9.9',
     ];
     for (const marker of need) {
       if (!out.includes(marker)) {

package/skills/adia-site-docs/evals/evals.json

@@ -0,0 +1,20 @@
+{
+  "skill": "adia-site-docs",
+  "note": "Trigger-routing suite (gh#369). expect=trigger: the skill should fire on the prompt in a fresh session. expect=no-trigger: a near-miss owned elsewhere (owner named per case) or generic knowledge with no AdiaUI skill involved. Cases derived from evals/routing-corpus.json's phrases array (gh#355 vetted corpus) plus the SKILL.md description's own trigger/NOT-fence language.",
+  "cases": [
+    { "id": "t01", "prompt": "Review this site docs page for consistency with the others", "expect": "trigger" },
+    { "id": "t02", "prompt": "This callout on the docs page reads as plain text, not a styled box — why?", "expect": "trigger" },
+    { "id": "t03", "prompt": "Add a new pattern page under site/pages/patterns for the kanban-board pattern", "expect": "trigger" },
+    { "id": "t04", "prompt": "Author a new getting-started guide page for theming dark mode", "expect": "trigger" },
+    { "id": "t05", "prompt": "Write a new architecture explainer page for the docs site", "expect": "trigger" },
+    { "id": "t06", "prompt": "Create a reference page documenting the MCP tools", "expect": "trigger" },
+    { "id": "t07", "prompt": "Audit site/pages/guides for the same callout bug we found elsewhere", "expect": "trigger" },
+    { "id": "t08", "prompt": "What's the shared skeleton every site docs page should follow?", "expect": "trigger" },
+    { "id": "n01", "prompt": "Review this component's own demo gallery in its .examples.html file", "expect": "no-trigger", "owner": "adia-author" },
+    { "id": "n02", "prompt": "Fix this gen-ui training exemplar page under site/pages/examples", "expect": "no-trigger", "owner": "adia-a2ui" },
+    { "id": "n03", "prompt": "Tune the zettel retrieval threshold for the chunk corpus", "expect": "no-trigger", "owner": "adia-a2ui" },
+    { "id": "n04", "prompt": "Add a new prop to the button-ui primitive", "expect": "no-trigger", "owner": "adia-author" },
+    { "id": "n05", "prompt": "Cut the next npm release for the 9 @adia-ai packages", "expect": "no-trigger", "owner": "adia-release" },
+    { "id": "n06", "prompt": "Deploy the docs site to ui-kit.exe.xyz", "expect": "no-trigger", "owner": "adia-deploy" }
+  ]
+}