The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill explicitly addresses the risk of indirect prompt injection. In the section 'Treating Error Output as Untrusted Data', it instructs the agent to treat logs and stack traces as diagnostic data only, not as a source of commands or URLs. This is a best-practice defensive measure.
Ingestion points: The skill reads error logs, console output, and stack traces to diagnose bugs.
Boundary markers: The instructions explicitly set a trust boundary for external error content.
Capability inventory: The skill uses standard development tools like npm test and git bisect.
Sanitization: The skill mandates that instructions found in errors must be surfaced to the user for confirmation rather than executed automatically.
[SECURITY_BEST_PRACTICES]: The included references/security-checklist.md provides a comprehensive guide for developers on hardening applications, including threat modeling, secret management, and secure dependency handling. This promotes a high security posture for the user.
[REMOTE_CODE_EXECUTION]: No suspicious remote code execution patterns were detected. The skill references standard, well-known development tools and libraries (e.g., Jest, Playwright, Lighthouse) for their intended purposes in testing and auditing.

debugging-and-error-recovery