The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill allows the execution of arbitrary shell commands through its support for 'release hooks' (prepare_artifact, publish_artifact) defined in a local .releaserc.yml configuration file. While this is a standard extension mechanism for release tools, it creates a surface where a repository containing a malicious configuration file could execute untrusted code when the release skill is invoked.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically commit messages via git log and pull request metadata via gh pr view, to automatically generate changelogs in multiple languages. This represents an indirect prompt injection surface where malicious input in a commit or PR could influence the agent's logic.
Ingestion points: Git commit history (git log) and GitHub Pull Request data (gh pr view) processed in SKILL.md.
Boundary markers: The instructions do not specify explicit delimiters or 'ignore' instructions for the agent when processing commit logs or PR bodies.
Capability inventory: Extensive shell command capabilities (git, gh) and the ability to execute user-defined hooks as specified in SKILL.md.
Sanitization: The skill explicitly requires that release notes be written to a temporary file rather than inlined into shell commands, which is a significant mitigation against command injection via changelog content.
[COMMAND_EXECUTION]: The skill relies on extensive use of system CLI tools, including git and the GitHub CLI (gh), to manage repository state, create tags, and publish releases. It also manages temporary files to handle multi-line strings for these commands.

release-skills