The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a template for a shell hook (pretooluse-hook.py) and a corresponding test script (pretooluse-hook-test.py). The test script uses subprocess.run() to verify that the hook correctly intercepts and blocks dangerous commands such as rm -rf / or force pushes to the main branch. This is legitimate behavior for a security component's test suite.
[SAFE]: Static detectors flagged destructive system commands in templates/artifacts/gates/pretooluse-hook.py. These commands are part of a defensive deny-list provided as a template for maintainers to harden their repositories against accidental agent actions. They are not intended for execution by the skill itself.
[PROMPT_INJECTION]: The references/playbooks/instruction-surface.md file contains references to prompt injection patterns like 'ignore previous instructions'. These are documented as part of an adversarial lens used to audit other repositories for vulnerabilities and do not represent an attempt to subvert the safety of the analyzing agent.
[COMMAND_EXECUTION]: The skill includes check-instruction-surface.sh, a utility script used to ensure that harness-specific instruction files (like CLAUDE.md) remain correctly symlinked to the main AGENTS.md file, preventing configuration drift.

agent-readiness