The Agent Skills Directory

[METADATA_POISONING]: The GitHub Actions workflow templates include deceptive comments claiming version numbers that do not exist for official actions. This may lead users to believe they are using more advanced or patched versions than they actually are.
Evidence: templates/release.yml and SKILL.md claim actions/checkout@v7.0.1 and actions/setup-node@v6.5.0 (actual versions are v4.x).
Evidence: SKILL.md claims dorny/paths-filter@v4.0.3 for a commit hash that corresponds to v3.0.0.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from commit messages and Pull Request titles to determine release logic and generate documentation.
Ingestion points: Conventional commit messages and PR titles are ingested via semantic-release (referenced in SKILL.md).
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are provided in the configuration templates.
Capability inventory: The skill has the capability to write to the repository (CHANGELOG.md, package.json), create GitHub Releases, and publish packages to npm via configured plugins (e.g., @semantic-release/git, @semantic-release/github, @semantic-release/npm).
Sanitization: The skill relies on the internal sanitization and parsing logic of the semantic-release plugins.
[COMMAND_EXECUTION]: The skill templates facilitate the execution of shell commands and inline Node.js scripts during the release process.
Evidence: templates/release.yml executes npx semantic-release to run the release pipeline.
Evidence: templates/releaserc.monorepo-package.json uses the @semantic-release/exec plugin to execute a node -e command that modifies package.json at runtime.

semantic-release-automation