Merge upstream main into fix/issue-283-color-scheme-echo

.coderabbit.yaml

@@ -0,0 +1,20 @@
+# yaml-language-server: $schema=https://www.coderabbit.ai/schema/schema.v2.json
+language: en-US
+reviews:
+  profile: chill
+  request_changes_workflow: false
+  auto_review:
+    enabled: false
+    auto_incremental_review: false
+    drafts: false
+    ignore_usernames:
+      - Kuddev
+  path_instructions:
+    - path: "**/*"
+      instructions: >-
+        Follow AGENTS.md and the closest module instructions. Prioritize actionable
+        correctness, regression, security, state/lifetime and compatibility issues.
+        Give the triggering scenario and concrete code evidence. Distinguish
+        confirmed defects from questions; avoid speculative redesign and formatting
+        comments. CI and maintainer review remain the merge constraints. Never
+        recommend weakening required checks, raising budgets, or removing tests.

.github/copilot-instructions.md

@@ -0,0 +1,19 @@
+# Review guidance
+
+Review concrete regressions and missing boundary cases in the changed code. Explain
+an actionable finding with the trigger, affected behavior, and file/line evidence.
+Distinguish confirmed defects from questions; avoid speculative redesigns and style
+comments already covered by formatting. Do not infer success from a passing build.
+
+Follow `AGENTS.md` and the closest module instructions, `CONTRIBUTING.md`,
+`docs/architecture.md`, and `docs/project-constraints.md`. Treat them as the
+source of project policy rather than inventing a second rule set. Check ownership,
+dependency direction, cancellation/lifetime, persistence compatibility, and tests
+appropriate to the change. Consult `docs/internationalization.md` for UI text and
+`packaging/AGENTS.md` for release changes. Verify performance claims against their
+stated workload and separate platform tests from visual acceptance.
+
+CI remains the deterministic merge gate. Do not recommend weakening required
+checks, widening budgets, deleting tests, or treating this automated review as
+maintainer approval. For CI changes, look for missing-platform, skipped-test,
+stale-commit, cache-poisoning, fork-permission, and failure-propagation regressions.

.github/workflows/copilot-review.yml

@@ -0,0 +1,147 @@
+name: AI review after CI
+
+on:
+  workflow_run:
+    workflows: [Full native tests, Architecture contracts, PR size]
+    types: [completed]
+  # Metadata only: never check out or execute pull-request content with this token.
+  pull_request_target:
+    types: [ready_for_review]
+  workflow_dispatch:
+    inputs:
+      pull_number:
+        description: Existing PR to evaluate with the same eligibility and CI checks
+        required: true
+        type: string
+
+permissions:
+  actions: read
+  checks: read
+  contents: read
+  issues: write
+  pull-requests: write
+
+# A pending event may be replaced. Every run scans eligible open PRs, so the
+# retained event still evaluates all PRs, and only one request writer runs at once.
+concurrency:
+  group: ai-first-review
+  cancel-in-progress: false
+
+jobs:
+  request-review:
+    if: github.event_name != 'workflow_run' || github.event.workflow_run.conclusion == 'success'
+    runs-on: ubuntu-24.04
+    timeout-minutes: 10
+    steps:
+      - uses: actions/github-script@v7
+        env:
+          COPILOT_REVIEW_TOKEN: ${{ secrets.COPILOT_REVIEW_TOKEN }}
+          CODERABBIT_FALLBACK_ENABLED: ${{ vars.CODERABBIT_FALLBACK_ENABLED }}
+        with:
+          retries: 0
+          script: |
+            const marker = 'ai-review-requested';
+            const previousMarker = 'copilot-review-requested';
+            const bot = 'copilot-pull-request-reviewer[bot]';
+            const fallbackBot = 'coderabbitai[bot]';
+            const same = (a, b) => a?.toLowerCase() === b?.toLowerCase();
+            const report = (number, text) => core.info(`#${number}: ${text}`);
+            const eligible = pr => pr.state === 'open' && pr.draft === false &&
+              pr.user?.type === 'User' && !same(pr.user.login, context.repo.owner);
+            let pulls;
+            if (context.eventName === 'workflow_dispatch') {
+              const value = context.payload.inputs?.pull_number;
+              if (!/^[1-9][0-9]*$/.test(value || '')) throw new Error('Invalid PR number');
+              pulls = [(await github.rest.pulls.get({
+                ...context.repo, pull_number: Number(value),
+              })).data];
+            } else {
+              pulls = await github.paginate(github.rest.pulls.list, {
+                ...context.repo, state: 'open', per_page: 100,
+              });
+            }
+            const policies = new Map();
+            let reviewer;
+            const provider = async () => {
+              if (!reviewer) {
+                const token = process.env.COPILOT_REVIEW_TOKEN;
+                if (!token) throw new Error('Missing dedicated COPILOT_REVIEW_TOKEN secret');
+                reviewer = new github.constructor({auth: token, retry: {enabled: false}});
+                const user = (await reviewer.rest.users.getAuthenticated()).data;
+                if (!same(user.login, context.repo.owner)) throw new Error('Review token must belong to the repository owner');
+              }
+              // This account endpoint is also used by GitHub's Copilot clients.
+              // Unknown or changed responses stop requests; never infer exhaustion
+              // from an authentication failure, a rate limit, or a network error.
+              const data = (await reviewer.request('GET /copilot_internal/user')).data;
+              const quota = data.quota_snapshots?.premium_interactions;
+              if (!quota || typeof quota.unlimited !== 'boolean' ||
+                  !Number.isFinite(quota.quota_remaining) || quota.quota_remaining < 0 ||
+                  !Number.isFinite(Date.parse(quota.timestamp_utc)) ||
+                  Math.abs(Date.now() - Date.parse(quota.timestamp_utc)) > 5 * 60 * 1000) {
+                throw new Error('Copilot allowance is missing, stale or unrecognized; no review requested');
+              }
+              if (quota.overage_permitted !== false) {
+                throw new Error('Paid overage must remain disabled before automated requests');
+              }
+              core.info(`Copilot allowance remaining: ${quota.quota_remaining}; reset: ${data.quota_reset_date_utc}`);
+              if (quota.unlimited || quota.quota_remaining > 0) return 'copilot';
+              if (process.env.CODERABBIT_FALLBACK_ENABLED !== 'true') {
+                throw new Error('Copilot exhausted; CodeRabbit installation must be verified before enabling fallback');
+              }
+              return 'coderabbit';
+            };
+            for (const pr of pulls) {
+              const number = pr.number;
+              if (!eligible(pr)) { report(number, 'Owner, bot, draft or closed PR; skipped.'); continue; }
+              const args = {...context.repo, pull_number: number};
+              const claimed = item => item.labels.some(label => [marker, previousMarker].includes(label.name)) ||
+                item.requested_reviewers.some(user => same(user.login, bot));
+              if (claimed(pr)) { report(number, 'Review already requested; skipped.'); continue; }
+              const reviews = await github.paginate(github.rest.pulls.listReviews, {...args, per_page: 100});
+              if (reviews.some(review => [bot, fallbackBot].some(login => same(review.user?.login, login)))) {
+                report(number, 'An AI reviewer has already reviewed this PR; manual re-review only.'); continue;
+              }
+              if (!policies.has(pr.base.ref)) {
+                const rules = await github.paginate('GET /repos/{owner}/{repo}/rules/branches/{branch}', {
+                  ...context.repo, branch: pr.base.ref, per_page: 100,
+                });
+                policies.set(pr.base.ref, rules.filter(rule => rule.type === 'required_status_checks')
+                  .flatMap(rule => rule.parameters.required_status_checks));
+              }
+              const required = policies.get(pr.base.ref);
+              if (!required.length) { report(number, 'No required CI policy; manual review only.'); continue; }
+              const ref = pr.head.sha;
+              const checks = await github.paginate(github.rest.checks.listForRef, {
+                ...context.repo, ref, filter: 'latest', per_page: 100,
+              });
+              const passed = required.every(rule => {
+                const matching = checks.filter(check => check.name === rule.context &&
+                  (rule.integration_id == null || rule.integration_id === -1 || check.app?.id === rule.integration_id));
+                matching.sort((a, b) => b.id - a.id);
+                return matching[0]?.head_sha === ref && matching[0]?.status === 'completed' &&
+                  matching[0]?.conclusion === 'success';
+              });
+              if (!passed) { report(number, 'Required CI incomplete or unsuccessful; deferred.'); continue; }
+              const selected = await provider();
+              // Guard a push, draft conversion or close racing the checks above.
+              const current = (await github.rest.pulls.get(args)).data;
+              if (!eligible(current) || current.head.sha !== ref || current.base.ref !== pr.base.ref || claimed(current)) {
+                report(number, 'PR changed while checking; deferred.'); continue;
+              }
+              // Persist the one-shot claim before the non-idempotent review request.
+              // On a timeout/error leave the claim in place: a maintainer must inspect
+              // the result before manually re-requesting, preventing quota retry loops.
+              await github.rest.issues.addLabels({
+                ...context.repo, issue_number: number, labels: [marker, `${selected}-review-requested`],
+              });
+              if (selected === 'copilot') {
+                await reviewer.rest.pulls.requestReviewers({...args, reviewers: [bot]});
+              } else {
+                await reviewer.rest.issues.createComment({
+                  ...context.repo, issue_number: number,
+                  body: '@coderabbitai review',
+                });
+              }
+              report(number, `Requested first ${selected} review after ${required.length} required checks passed.`);
+            }

.github/workflows/linux-lua.yml

@@ -63,7 +63,8 @@ jobs:
             scripts.tests.test_stable_release \
             scripts.tests.test_ci_cache \
             scripts.tests.test_platform_cfg \
-            scripts.tests.test_pr_size_workflow -v
+            scripts.tests.test_pr_size_workflow \
+            scripts.tests.test_copilot_review_workflow -v
       - name: Check platform cfg budget before native jobs
         run: python3 scripts/check_platform_cfg.py
       - name: Plan native validation

architecture/notes/scripts/ci/2026-09-28-copilot-after-required-ci.md

@@ -0,0 +1,98 @@
+# AI review after required CI, with an allowance fallback
+
+## Status
+
+Implemented; activation requires the default-branch workflow, a dedicated owner
+credential, and verified CodeRabbit installation before enabling its fallback.
+
+## Context
+
+Reviews of every push spend allowance on drafts, failed builds, and changes that
+already have an initial review. Native branch rules do not express the requested
+combination of authorship, readiness, CI, review history, and allowance routing.
+
+## Evidence
+
+GitHub exposes effective required checks through the branch rules API and results
+for a specific commit. The reviewer API accepts `copilot-pull-request-reviewer[bot]`.
+A user-authenticated request produced a review, whereas an Actions-token probe
+returned success without a Copilot run or review. API acceptance alone therefore
+cannot establish delivery. A user credential is required for activation testing.
+
+The authenticated Copilot client endpoint exposes a timestamped allowance snapshot,
+including remaining credits and whether paid overage is permitted. This endpoint
+is not a stable public billing contract; response drift must stop automation.
+
+## Decision
+
+The metadata-only [workflow](../../../../.github/workflows/copilot-review.yml)
+requests one review for an open, non-draft PR by a human other than the repository
+owner, after every required check on its current head succeeds. It reads effective
+server-side branch rules rather than copying their check list. No required-check
+policy means no automatic review.
+
+Completed CI workflows and draft-to-ready transitions trigger eligibility scans.
+A repository-wide concurrency group serializes requests; scanning open PRs covers
+events replaced in GitHub's pending slot. Dispatch evaluates one existing PR with
+the same guards. Previous Copilot or CodeRabbit reviews prevent automatic repeats.
+
+The workflow's built-in token reads PR/CI metadata and writes claim labels.
+`COPILOT_REVIEW_TOKEN` must belong to the repository owner and is used only for
+account allowance and user-attributed review requests/comments; it needs no Checks,
+Actions or Contents permission for metadata reads. A positive Copilot
+allowance selects Copilot. Zero allowance selects CodeRabbit only when
+`CODERABBIT_FALLBACK_ENABLED=true`, after installation has been verified. Missing,
+negative, stale, unrecognized, or unreadable allowance data stops requests; an
+ordinary permission/network error never implies exhaustion. Paid overage must be
+disabled. Credits are not a fixed number of reviews, and consumption reporting can
+lag; this router does not establish a provider-side spending cap.
+
+A persistent `ai-review-requested` label is written before requesting either bot;
+older `copilot-review-requested` claims remain recognized. On uncertain delivery or
+failure, keep the claim and require manual inspection before retrying. No automatic
+retry loops or default dual review. CodeRabbit's own automatic and incremental
+reviews are disabled in the repository config; the fallback uses an explicit
+`@coderabbitai review` request. Instructions prioritize evidence-backed defects.
+
+The workflow never checks out PR code, runs artifacts, interpolates PR text into
+commands, approves a PR, or merges changes. Human review and required CI remain
+unchanged. The dedicated credential is a repository secret, never checked in;
+provision and validate it separately from the workflow's source.
+
+## Rejected alternatives
+
+- Branch-wide rules or every-push reviews: cannot enforce the requested policy.
+- Executing fork code with privileged credentials: unnecessary for metadata routing.
+- Treating any Copilot error as exhausted credit: hides outages and permission bugs.
+- In-memory deduplication or uncertain retries: can request two paid reviews.
+- Hard-coded check names: drift from the effective merge requirements.
+- Copying a developer's broad CLI credential into Actions without explicit setup.
+
+## Consequences
+
+Maintainers request subsequent reviews manually. A push can race the final API
+call; rechecking head/draft/base/claim immediately before submission narrows but
+cannot eliminate that race. An accepted request followed by asynchronous service
+failure remains for manual inspection rather than risking duplicate reviews.
+Unprotected target branches and legacy status-only checks require manual review.
+This is supplemental automation, not another mandatory merge check.
+
+## Validation
+
+Offline fixtures execute the actual script: successful external PR; owner, bot,
+draft and closed exclusions; missing/failed/skipped checks; newer pending attempts;
+wrong app/head; racing changes; review history; claims; policy/quota failures;
+positive versus exhausted allowance; disabled fallback; and no double request.
+The required lint job runs the contract before scheduling native runners.
+Actual owner-secret delivery and CodeRabbit response must be verified on GitHub
+before claiming end-to-end activation. Do not consume the real allowance merely
+to force exhaustion; the zero-credit routing boundary is covered with fixtures.
+
+## Supersedes
+
+None.
+
+## Revisit when
+
+GitHub exposes a stable allowance API or equivalent native routing, supports atomic
+SHA-bound review requests, required-check providers change, or ownership changes.

scripts/tests/test_copilot_review_workflow.py

@@ -0,0 +1,243 @@
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from pathlib import Path
+import shutil
+import subprocess
+import textwrap
+import unittest
+
+ROOT = Path(__file__).resolve().parents[2]
+RUNNER = r'''
+const input = JSON.parse(require('node:fs').readFileSync(0, 'utf8'));
+const writes = [], logs = [];
+process.env.COPILOT_REVIEW_TOKEN = input.missingToken ? '' : 'fixture-token';
+process.env.CODERABBIT_FALLBACK_ENABLED = input.fallbackEnabled ? 'true' : 'false';
+const list = () => {}, listReviews = () => {}, listForRef = () => {};
+const github = {
+  rest: {
+    pulls: {
+      list, listReviews,
+      get: async () => ({data: input.current || input.pr}),
+    },
+    checks: {listForRef},
+    issues: {addLabels: async value => {
+      if (input.labelError) throw new Error('claim failed');
+      writes.push(['claim', value]);
+    }},
+  },
+  paginate: async (method, args) => {
+    if (method === list) return [input.pr];
+    if (method === listReviews) return input.reviews || [];
+    if (method === listForRef) return input.checks;
+    if (typeof method === 'string' && method.includes('/rules/branches/')) {
+      if (input.policyError) throw new Error('policy unavailable');
+      return input.rules;
+    }
+    throw new Error('unexpected API');
+  },
+};
+const reviewer = {
+  rest: {
+    users: {getAuthenticated: async () => ({data: {login: input.tokenOwner || 'owner'}})},
+    pulls: {requestReviewers: async value => {
+      writes.push(['request', value]);
+      if (input.requestError) throw new Error('request failed');
+    }},
+    issues: {createComment: async value => writes.push(['fallback', value])},
+  },
+  request: async () => {
+    if (input.quotaError) throw new Error('quota API failed');
+    return {data: {quota_snapshots: {premium_interactions: input.quota || {
+      unlimited: false, quota_remaining: 100, overage_permitted: false,
+      timestamp_utc: new Date().toISOString(),
+    }}}};
+  },
+};
+github.constructor = class {
+  constructor({auth}) {
+    if (auth !== 'fixture-token') throw new Error('Wrong credential');
+    return reviewer;
+  }
+};
+const context = {repo: {owner: 'owner', repo: 'project'}, eventName: input.event || 'workflow_run',
+  payload: {inputs: {pull_number: input.number || '7'}}};
+const core = {info: value => logs.push(value)};
+const AsyncFunction = Object.getPrototypeOf(async function () {}).constructor;
+new AsyncFunction('github', 'context', 'core', input.script)(github, context, core)
+  .then(() => process.stdout.write(JSON.stringify({writes, logs})))
+  .catch(error => process.stdout.write(JSON.stringify({writes, logs, error: error.message})));
+'''
+
+
+class CopilotReviewWorkflowTests(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.node = shutil.which('node')
+        if not cls.node:
+            raise RuntimeError('Node.js is required for the Copilot workflow contract')
+        cls.workflow = (ROOT / '.github/workflows/copilot-review.yml').read_text(encoding='utf-8')
+        cls.script = textwrap.dedent(cls.workflow.split('          script: |\n', 1)[1])
+
+    def fixture(self):
+        return {
+            'pr': {'number': 7, 'state': 'open', 'draft': False,
+                   'user': {'login': 'contributor', 'type': 'User'},
+                   'head': {'sha': 'current'}, 'base': {'ref': 'main'},
+                   'labels': [], 'requested_reviewers': []},
+            'rules': [{'type': 'required_status_checks', 'parameters': {
+                'required_status_checks': [{'context': 'native', 'integration_id': 15368},
+                                           {'context': 'architecture', 'integration_id': 15368}]}}],
+            'checks': [{'id': i, 'name': name, 'head_sha': 'current', 'status': 'completed',
+                        'conclusion': 'success', 'app': {'id': 15368}}
+                       for i, name in enumerate(('native', 'architecture'), 1)],
+        }
+
+    def run_gate(self, data):
+        result = subprocess.run([self.node, '-e', RUNNER],
+                                input=json.dumps({'script': self.script, **data}),
+                                text=True, encoding='utf-8', capture_output=True, check=True)
+        return json.loads(result.stdout)
+
+    def test_external_ready_pr_claimed_then_requested_after_all_required_ci(self):
+        result = self.run_gate(self.fixture())
+        self.assertEqual([row[0] for row in result['writes']], ['claim', 'request'])
+        self.assertEqual(result['writes'][1][1]['reviewers'], ['copilot-pull-request-reviewer[bot]'])
+
+    def test_owner_draft_closed_and_bot_prs_never_request(self):
+        for change in ({'user': {'login': 'OwNeR', 'type': 'User'}}, {'draft': True},
+                       {'state': 'closed'}, {'user': {'login': 'dependabot', 'type': 'Bot'}}):
+            with self.subTest(change=change):
+                data = self.fixture()
+                data['pr'].update(change)
+                self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_failed_pending_skipped_neutral_missing_wrong_app_and_old_checks_defer(self):
+        for change in ({'conclusion': 'failure'}, {'status': 'in_progress'},
+                       {'conclusion': 'skipped'}, {'conclusion': 'neutral'},
+                       {'name': 'unrelated'}, {'app': {'id': 123}}, {'head_sha': 'old'}):
+            with self.subTest(change=change):
+                data = self.fixture()
+                data['checks'][0].update(change)
+                self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_newer_pending_attempt_cannot_reuse_old_success(self):
+        data = self.fixture()
+        data['checks'].append({**data['checks'][0], 'id': 99, 'status': 'queued', 'conclusion': None})
+        self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_previous_review_on_any_commit_prevents_automatic_rereview(self):
+        data = self.fixture()
+        data['reviews'] = [{'user': {'login': 'copilot-pull-request-reviewer[bot]'}, 'commit_id': 'old'}]
+        self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_pending_request_or_persistent_claim_prevents_duplicates(self):
+        for field, value in (
+            ('labels', [{'name': 'copilot-review-requested'}]),
+            ('requested_reviewers', [{'login': 'copilot-pull-request-reviewer[bot]'}]),
+        ):
+            data = self.fixture()
+            data['pr'][field] = value
+            self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_push_draft_close_base_change_or_claim_during_reads_defers(self):
+        for change in ({'head': {'sha': 'new'}}, {'draft': True}, {'state': 'closed'},
+                       {'base': {'ref': 'release'}}, {'labels': [{'name': 'copilot-review-requested'}]}):
+            with self.subTest(change=change):
+                data = self.fixture()
+                data['current'] = {**data['pr'], **change}
+                self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_missing_or_unreadable_policy_fails_closed(self):
+        data = self.fixture()
+        data['rules'] = []
+        self.assertEqual(self.run_gate(data)['writes'], [])
+        data['policyError'] = True
+        self.assertEqual(self.run_gate(data)['error'], 'policy unavailable')
+
+    def test_write_failure_does_not_retry_or_erase_claim(self):
+        data = self.fixture()
+        data['requestError'] = True
+        result = self.run_gate(data)
+        self.assertEqual(result['error'], 'request failed')
+        self.assertEqual([row[0] for row in result['writes']], ['claim', 'request'])
+        data['labelError'] = True
+        self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_manual_backfill_uses_same_gate_and_validates_number(self):
+        data = {**self.fixture(), 'event': 'workflow_dispatch'}
+        self.assertEqual(len(self.run_gate(data)['writes']), 2)
+        data['number'] = '7; arbitrary code'
+        self.assertEqual(self.run_gate(data)['error'], 'Invalid PR number')
+
+    def test_quota_exhaustion_uses_only_enabled_coderabbit(self):
+        data = self.fixture()
+        data['quota'] = {'unlimited': False, 'quota_remaining': 0, 'overage_permitted': False,
+                         'timestamp_utc': datetime.now(timezone.utc).isoformat()}
+        self.assertEqual(self.run_gate(data)['writes'], [])
+        data['fallbackEnabled'] = True
+        result = self.run_gate(data)
+        self.assertEqual([row[0] for row in result['writes']], ['claim', 'fallback'])
+        self.assertEqual(result['writes'][1][1]['body'], '@coderabbitai review')
+
+    def test_quota_errors_stale_data_paid_overage_and_wrong_identity_fail_closed(self):
+        for extra in ({'quotaError': True}, {'tokenOwner': 'other'},
+                      {'quota': {'unlimited': False, 'quota_remaining': -1}},
+                      {'quota': {'unlimited': False, 'quota_remaining': 10,
+                                 'overage_permitted': True, 'timestamp_utc': '2020-01-01T00:00:00Z'}}):
+            with self.subTest(extra=extra):
+                data = {**self.fixture(), **extra, 'fallbackEnabled': True}
+                result = self.run_gate(data)
+                self.assertEqual(result['writes'], [])
+                self.assertIn('error', result)
+
+    def test_paid_overage_or_stale_allowance_never_routes_to_either_bot(self):
+        for quota in (
+            {'unlimited': False, 'quota_remaining': 50, 'overage_permitted': True,
+             'timestamp_utc': datetime.now(timezone.utc).isoformat()},
+            {'unlimited': False, 'quota_remaining': 0, 'overage_permitted': False,
+             'timestamp_utc': '2020-01-01T00:00:00Z'},
+        ):
+            data = {**self.fixture(), 'quota': quota, 'fallbackEnabled': True}
+            self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_missing_user_token_does_not_spend_or_claim(self):
+        result = self.run_gate({**self.fixture(), 'missingToken': True})
+        self.assertEqual(result['writes'], [])
+        self.assertIn('Missing dedicated', result['error'])
+
+    def test_user_token_has_no_ci_reading_methods(self):
+        # The two clients deliberately expose disjoint APIs in the fixture.
+        result = self.run_gate(self.fixture())
+        self.assertEqual([row[0] for row in result['writes']], ['claim', 'request'])
+        self.assertNotIn('github-token: ${{ secrets.', self.workflow)
+
+    def test_available_copilot_does_not_trigger_enabled_fallback(self):
+        result = self.run_gate({**self.fixture(), 'fallbackEnabled': True})
+        self.assertEqual([row[0] for row in result['writes']], ['claim', 'request'])
+
+    def test_coderabbit_config_disables_independent_automatic_reviews(self):
+        config = (ROOT / '.coderabbit.yaml').read_text(encoding='utf-8')
+        for entry in ('enabled: false', 'auto_incremental_review: false', 'drafts: false', '- Kuddev'):
+            self.assertIn(entry, config)
+
+    def test_coderabbit_does_not_duplicate_previous_ai_review(self):
+        data = self.fixture()
+        data['reviews'] = [{'user': {'login': 'coderabbitai[bot]'}}]
+        self.assertEqual(self.run_gate(data)['writes'], [])
+
+    def test_workflow_never_executes_fork_content_or_approves_merges(self):
+        self.assertNotIn('actions/checkout', self.workflow)
+        self.assertIn('secrets.COPILOT_REVIEW_TOKEN', self.workflow)
+        self.assertNotIn('contents: write', self.workflow)
+        self.assertNotIn('createReview', self.script)
+        self.assertNotIn('synchronize', self.workflow)
+        self.assertIn('types: [ready_for_review]', self.workflow)
+        self.assertIn("github.event.workflow_run.conclusion == 'success'", self.workflow)
+        self.assertIn('retries: 0', self.workflow)
+        self.assertIn('cancel-in-progress: false', self.workflow)
+
+
+if __name__ == '__main__':
+    unittest.main()