The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous commands for setting up development environments and release workflows. These include using npm and pip for package management, npx for tool execution, and direct shell commands for configuration (cat, printf, git config). These are standard for the described purpose of changelog automation.
[EXTERNAL_DOWNLOADS]: The implementation guide references several external dependencies from official registries. It installs packages such as semantic-release, standard-version, and commitizen from npm and PyPI. It also utilizes official GitHub Actions from the actions organization (actions/checkout, actions/setup-node) for CI/CD workflows.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process commit messages and Pull Request descriptions, which are potentially attacker-controlled in open-source or multi-contributor environments.
Ingestion points: Commit messages, PR titles, and descriptions are read by tools like semantic-release and git-cliff to generate changelogs (resources/implementation-playbook.md).
Boundary markers: The instructions do not provide specific boundary markers or system prompts to isolate these inputs from the LLM's instructional context if it is used to summarize them.
Capability inventory: The skill has the capability to write to the file system (CHANGELOG.md, package.json), tag repositories, and publish to external registries like npm and GitHub.
Sanitization: The skill includes a mitigation step in SKILL.md advising the user to "Review output for accuracy, completeness, and wording" before publication.
[DYNAMIC_EXECUTION]: The skill implements git hooks using husky (resources/implementation-playbook.md), which involves executing shell scripts (commit-msg hook) automatically during git operations. It also utilizes semantic-release plugins which can execute custom logic during the release pipeline.

changelog-automation