The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it processes untrusted repository data to generate pull request content.
Ingestion points: Branch diffs, git logs, and configuration files such as AGENTS.md or CLAUDE.md (SKILL.md).
Boundary markers: None explicitly defined for isolating code diffs or commit messages.
Capability inventory: Capability to perform repository searches using grep and write permissions to pull-requests (SKILL.md).
Sanitization: No specific sanitization methods are identified for the ingested code or log data. The inherent risk is mitigated by instructions that require the agent to verify the output against a specific goal and to include a 'Ship checklist' designed for human oversight of sensitive changes.
[COMMAND_EXECUTION]: The skill involves the use of searching tools like grep to track modifications to project guardrails and rules within the codebase (SKILL.md). This command execution is localized and consistent with the skill's intended purpose of ensuring documentation and code integrity during the pull request process.

file-pr