The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs repository operations using standard tools including git and the GitHub CLI (gh). These commands are used to fetch tags, verify the working tree state, and create new releases on GitHub.
Evidence: Steps 1, 2, and 6 use commands such as git fetch origin --tags, git tag -l, gh release view, and gh release create.
[DYNAMIC_EXECUTION]: The skill dynamically executes build and test scripts detected from project manifests or configuration files. This includes running arbitrary commands specified in project-specific documentation.
Evidence: Step 3 defines a resolution order for build checks, explicitly stating it will "run exactly that" for commands found in the ## Releasing section of CLAUDE.md or AGENTS.md. It also executes scripts from package.json and composer.json using detected package managers like bun, pnpm, yarn, or npm.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the target project which could be manipulated to influence the agent's actions, particularly during the build and changelog generation phases.
Ingestion points: The skill reads project configuration files (CLAUDE.md, AGENTS.md) and fetches git tag metadata from the repository (Steps 1, 3, and 4).
Boundary markers: A mandatory review gate is implemented in Step 5, requiring explicit user approval before the final release creation. However, the build and test commands in Step 3 are executed automatically if detected.
Capability inventory: The agent has the ability to execute shell commands, perform network operations via GitHub CLI, and write temporary files to the <scratchpad> directory.
Sanitization: The skill uses the --notes-file parameter when calling gh release create to prevent shell injection from the changelog content.

release