The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository sources including commit messages, branch names, and pull request templates. This content is interpolated into the final PR title and body, creating a surface for indirect instructions to potentially influence the agent's behavior.
Ingestion points: git log output, git diff output, and various pull/merge request template files (e.g., .github/pull_request_template.md or .gitlab/merge_request_templates/*.md).
Boundary markers: Content is organized under markdown headers (e.g., ## Summary), but specific escaping for untrusted interpolation is not defined.
Capability inventory: Pull request creation and editing via gh and glab CLIs, and git push operations.
Sanitization: The skill implements a "No secrets in the body" guardrail, scanning for .env files, keys, and tokens to exclude them. It also strips agent attribution lines.
[COMMAND_EXECUTION]: The skill makes extensive use of local CLI tools such as git, gh, glab, and jq. It follows a least-privilege model by only pre-approving read-only commands in its allowed-tools configuration, ensuring that actions that modify state (like creating a PR or pushing code) require explicit user confirmation.
[SAFE]: The skill references a configuration schema hosted on the author's own GitHub repository (TitusKirch/skills). This is a legitimate vendor resource used for configuration validation and does not represent a security risk.

pull-request