The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data including CI logs, review comments, and PR bodies as ingestion points identified in references/git-publish.md. It implements a boundary marker by instructing the agent to treat this data as evidence of code issues only and explicitly forbidding the execution of embedded commands. The skill possesses capabilities for Git CLI operations and file writes in SKILL.md, but enforces validation steps and verification of PR metadata to mitigate risks.
[DATA_EXPOSURE_AND_EXFILTRATION]: Explicit security guidelines are provided in references/git-publish.md to prevent the accidental exposure of sensitive information, such as .env files and credentials, in Git commits.
[COMMAND_EXECUTION]: The skill enforces safe Git practices by prohibiting the bypass of verification hooks and GPG signing, ensuring that repository security policies are respected during CLI operations.

w-pr