The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the gh CLI and tmux for automated triage and testing. To mitigate command injection, it implements a strict sanitization pattern using tr -cd '[:alnum:] _-.,' to filter all but safe characters from user-provided scenario strings before they are passed to shell keys.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from GitHub issues and pull requests. It manages this risk through a mandatory "Untrusted input" rule that forbids the interpolation of issue/PR text directly into shell commands. It also requires the use of enter_worktree to ensure all file-system operations are performed within a scoped, temporary directory rather than the main project root.
[SAFE]: The skill contains an explicit security guardrail that forbids the execution of PR-derived code in CI environments. This prevents a common attack vector where a malicious pull request might attempt to exfiltrate the CI environment's write-access tokens (PAT) by triggering a build or test run of modified scripts.

triage