The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests potentially untrusted data from the repository's git commit history and diffs to generate summaries for the CHANGELOG.md file. This creates an attack surface where malicious instructions could be embedded in commit messages to influence the agent's behavior during the release workflow.
Ingestion points: Git commit logs and repository diffs used for changelog generation in SKILL.md.
Capability inventory: Shell command execution (git, pnpm) and file system modifications (package.json, CHANGELOG.md) as listed in references/release-checklist.md.
Boundary markers: No specific delimiters or safety instructions are provided to the agent to distinguish between data and instructions when reading commit history.
Sanitization: The instructions do not require the agent to sanitize or validate the content of commit messages before processing them.
[COMMAND_EXECUTION]: The skill provides a comprehensive set of shell commands for the agent to execute, including git branch management, tag creation, and running repository-defined scripts via pnpm (such as pnpm run lint and pnpm run release:ff). While these commands are standard for release automation, they grant the agent significant control over the local and remote repository state.

deepchat-release