.gitea/workflows/int9-guard.yaml

@@ -0,0 +1,41 @@
+name: INV-INT9 issue.status write-ownership
+
+# Hard gate for INV-INT9 (legal-ai docs/spec/X7-paperclip-client-params.md
+# §4): `issue.status` has exactly one legitimate writer — the
+# `sync-case-status` scheduled job (`ctx.jobs.register("sync-case-status", …)`
+# in src/worker.ts), which mirrors legal-ai's case status onto the linked
+# Paperclip issue. Built after legal-ai issue #446, where a second, unnoticed
+# writer flip-flopped an issue `done → in_progress` — nobody caught it because
+# nothing enforced write-ownership. This gate makes sure a third writer never
+# lands the same way again: any `ctx.issues.update(...)` call that touches
+# `status` outside the `sync-case-status` job fails the build, unless it
+# carries an explicit, justified `// noqa: INT9 — <reason>` (a bare noqa with
+# no reason is itself a violation — exceptions must be argued in writing).
+#
+# Pure Node, zero dependencies (scripts/int9-guard.mjs) — no `npm ci` needed.
+#
+# ⚠️ חסום למיזוג. אין היום runner ל-Gitea Actions בהיקף הריפו הזה: ה-runner
+# היחיד על nautilus רשום בהיקף-repo ל-`legal-ai` בלבד (`action_runner.repo_id = 6`),
+# ולכן job שנפתח כאן לא נאסף לעולם — נמדד ב-legal-ai #618: ריצה 3200 / job 3250
+# ישבה ב-`queued` עם `started_at: 1970-01-01` בזמן שה-runner היה `busy: false`.
+# ה-PR הזה נשאר פתוח **בכוונה** ומשמש כמדידה: ביום שיירשם runner לריפו הזה הוא
+# יהפוך לירוק מעצמו, וזו תהיה הראיה ש-AC2 של #618 סופק. עד אז השער רץ ידנית:
+# `npm run int9:guard` · `npm run int9:guard:self-test`.
+
+on:
+  pull_request:
+    branches: [main]
+  push:
+    branches: [main]
+
+jobs:
+  int9-guard:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: INV-INT9 — issue.status write-ownership guard
+        run: node scripts/int9-guard.mjs
+
+      - name: Self-test — הוכחה שהשער נושך (AC5)
+        run: node scripts/int9-guard.mjs --self-test