Add stop hook to prevent early stopping with incomplete tasks (#2325)

## Summary - Add a Stop hook that checks for incomplete tasks before allowing Claude to stop - The hook blocks stopping if any tasks are in pending or in_progress status - Includes comprehensive tests for the hook behavior ## Test plan - Run the test suite: python3 .claude/hooks/tests/test_stop_hook.py - Verify all tests pass (no tasks allows stop, completed tasks allow stop, pending/in_progress tasks block stop) #skip-bugbot Generated with Claude Code  --- ## Summary by cubic Adds a Stop hook to prevent early stopping when work isn’t done. Uses a stricter, prompt-based check of the conversation context with a default bias to continue. - **New Features** - Stop hook returns {"ok": true} to allow stop or {"ok": false, "reason": "..."} to continue. - Evaluates task completion, unresolved errors, follow-ups, and work quality. - Biases to continue unless everything is truly complete; registered under Stop in .claude/settings.json with a 30s timeout. <sup>Written for commit 059862cf76e0df9dd9149df2c8acd840c06dbc15. Summary will update on new commits.</sup>  --------- Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Add stop hook to prevent early stopping with incomplete tasks (#2325)
20938765 · Will Chen · GitHub · 01acdbf2 · 20938765
--- a/.claude/settings.json
+++ b/.claude/settings.json
@@ -147,6 +147,17 @@
          }
        ]
      }
+    ],
+    "Stop": [
+      {
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "You are evaluating whether Claude should stop working. Your default bias should be to CONTINUE WORKING (ok: false). Only allow stopping when absolutely certain all work is complete.\n\nContext: $ARGUMENTS\n\nAnalyze the conversation and determine:\n\n1. TASK COMPLETION CHECK\n   - Has EVERY task the user requested been fully completed?\n   - Not just acknowledged, but actually done with verifiable output?\n   - Are there any implied subtasks that haven't been addressed?\n\n2. ERROR CHECK\n   - Did any errors occur that weren't resolved?\n   - Are there failed operations that need retry?\n   - Did Claude say it would do something but didn't actually do it?\n\n3. FOLLOW-UP CHECK\n   - Is there obvious next-step work that should be done?\n   - Did the user's request imply multiple steps where only some were completed?\n   - Would a thorough assistant do more here?\n\n4. QUALITY CHECK\n   - Was the work done to completion, not just partially?\n   - Are there placeholders, TODOs, or incomplete sections?\n   - Did Claude stop prematurely mid-task?\n\nDECISION RULES:\n- If ANY task is incomplete → continue (ok: false)\n- If ANY error is unresolved → continue (ok: false)\n- If there's reasonable follow-up work → continue (ok: false)\n- If Claude said \"I'll do X\" but didn't → continue (ok: false)\n- If work quality is partial/rushed → continue (ok: false)\n- ONLY return ok: true if ALL tasks are genuinely, fully complete AND no errors exist AND no reasonable follow-up remains\n\nRespond with JSON:\n- {\"ok\": false, \"reason\": \"specific explanation of what still needs to be done\"} — DEFAULT, use this unless everything is truly complete\n- {\"ok\": true} — ONLY when absolutely certain all work is finished and nothing remains",
+            "timeout": 30
+          }
+        ]
+      }
    ]
  }
 }