kin: KIN-136-backend_dev

agents/prompts/reviewer.md

@@ -72,7 +72,8 @@ Example:
   "security_issues": [],
   "conventions_violations": [],
   "test_coverage": "adequate",
-  "summary": "Implementation looks correct and follows project patterns. One minor style issue noted."
+  "summary": "Implementation looks correct and follows project patterns. One minor style issue noted.",
+  "exit_condition": null
 }
 ```
 
@@ -100,6 +101,22 @@ Example:
     }
     ```
 
+**`exit_condition`** (optional, KIN-136 auto-return):
+
+Set this field ONLY when the task cannot be auto-retried and requires a human decision:
+
+- `"login_required"` — the reviewer or the code requires a login/auth that is not available in automation context
+- `"missing_data"` — critical data, credentials, or access needed to continue is missing and cannot be inferred
+- `"strategic_decision"` — the fix requires a fundamental architectural or business decision with no obvious correct answer (e.g. conflicting stakeholder requirements, irreversible platform choice)
+
+Leave as `null` in ALL other cases — including ordinary bugs, quality issues, missing tests, style violations, or any fixable problem. When `null`, the system will automatically retry the task with a return analyst.
+
+**When NOT to set exit_condition (set null):**
+- Code has bugs or logic errors → `null` (auto-retry will fix)
+- Tests are missing or failing → `null` (auto-retry will add tests)
+- Implementation doesn't match requirements → `null` (auto-retry will revise)
+- Security issue found → `null` with `"changes_requested"` verdict (auto-retry will patch)
+
 **`security_issues` and `conventions_violations`** elements:
 
 ```json