The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill interacts with GitHub (api.github.com) and GitLab (gitlab.com) APIs to fetch repository metadata, release information, and assets. These downloads target well-known, trusted service providers for their primary intended purpose.
[COMMAND_EXECUTION]: The instructions utilize the GitHub CLI (gh), GitLab CLI (glab), and standard utilities such as curl, jq, and sort to perform repository management tasks and process API responses.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process content from hosted repository resources (e.g., Pull Request descriptions, Issue bodies, and release notes) which could contain attacker-controlled data.
Ingestion points: Data enters the context via API responses from GitHub and GitLab providers, as described in SKILL.md and the reference files.
Boundary markers: The skill explicitly instructs the agent to treat pull-request and fork content as untrusted input, even when the transport itself is trusted (SKILL.md).
Capability inventory: The skill leverages shell command execution through gh and glab to perform repository reads and authorized mutations (references/github-api.md).
Sanitization: The documentation recommends the use of parameterized GraphQL variables and response schema validation to mitigate injection risks (references/auth-and-security.md).

git-actions