| name | integration-test-fixer |
| description | Automatically run and fix integration tests for coding agent configurations. |
Integration Test Fixer
This skill automates the process of running the container-based integration tests, analyzing failures, and proposing/applying fixes.
Purpose
To ensure that coding agent security configurations (hooks, sandboxes, etc.) remain valid and functional across all supported agents (Claude, Gemini, Cursor, Codex, Copilot).
Instructions
1. Run the Verification Script
Execute the helper script to run the tests and get a structured report:
./.claude/skills/integration-test-fixer/scripts/verify-and-fix.sh
2. Analyze the Report
- If the script returns success, the tests are passing.
- If it fails, examine the JSON output or the console log to identify the failing test case.
- Cross-reference the failure with references/failure-patterns.md.
3. Apply Fixes
- Path Issues: Check
integration_tests/Dockerfile or scripts/activate-configs.sh.
- Hook Logic: Check the respective
gh-safeguard.sh or agent-specific hook script.
- Dependency Issues: Ensure the
Dockerfile has all necessary libraries (e.g., libnss3 for Cursor).
4. Verify
Re-run the verification script after applying changes to ensure the regression is resolved.
References
