The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local shell scripts located in the .claude/hooks/ directory, specifically ci-wait.sh, lint.sh, and test.sh. This relies on the security of the local repository environment.
[COMMAND_EXECUTION]: Extensive use of the gh (GitHub CLI) for repository operations, including fetching logs (gh run view --log-failed), polling API endpoints, and pushing code changes to the remote repository.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via external data sources.
Ingestion points: Untrusted data enters the agent context through GitHub PR review bodies, line-level comments, and CI failure logs.
Boundary markers: Absent. The instructions do not specify any delimiters or safety warnings to ignore instructions embedded within the logs or comments.
Capability inventory: The skill possesses Bash, Edit, Write, and Skill (specifically invoking ralph-loop) capabilities, allowing it to modify source code and execute shell commands based on the ingested data.
Sanitization: None. The skill is directed to "Understand the suggestion" and "apply the fix" directly from PR comments, which could contain malicious instructions designed to exfiltrate data or compromise the build environment.
[COMMAND_EXECUTION]: The skill provides for autonomous code modification and pushing to a remote repository in response to CI failures without a mandatory human-in-the-loop verification (though a verification step exists for PR review fixes).

watch-pr