The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands to manage the development workflow:
git commit and git push for version control operations.
gh pr edit and gh api graphql to interact with GitHub pull requests.
task changelog to run project-specific tasks defined in a local Taskfile.yaml.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository data to generate pull request descriptions and changelog entries. This creates an attack surface where content in the codebase (e.g., diffs or commit messages) could influence the output generated by the subagent.
Ingestion points: File metadata (GIT_COMMIT_FULL), directory structures (.changeset/, .changes/), and output from git diff and commit logs in SKILL.md.
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the subagent prompt.
Capability inventory: Capabilities include network communication via git push and repository modification via gh pr edit and local file writes to /tmp/pr-body.md.
Sanitization: No explicit sanitization or validation of the processed repository content is performed before interpolation into the subagent prompt.

ship-it