The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Git commands (diff, rev-parse, ls-files, show) via a TypeScript script (scripts/changeset-plan.ts). Risk is mitigated by a strict whitelist for branch references (origin/dev, origin/minor, origin/major) in the parseBaseRef function, preventing arbitrary command injection through the --base-ref argument.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the repository's environment that could be influenced by external contributors.
Ingestion points: package.json manifests and Git diff outputs are read and parsed by scripts/changeset-plan.ts and provided to the agent.
Boundary markers: The skill explicitly instructs the agent to present a draft to the user and wait for confirmation before writing any files. It also instructs the agent not to automatically guess bumps or meanings of changes.
Capability inventory: The skill can write .changeset/*.md files to the local file system. It uses git for status and content retrieval but does not perform writes, commits, or pushes.
Sanitization: CLI arguments are validated; however, the agent directly interprets the content of package descriptions and commit logs to propose SemVer bumps.

seed-changeset