The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local environment by executing git commands and forge-specific CLI tools like gh (GitHub) and fj (Forgejo). It includes specific instructions to use safe practices, such as using --body-file for multiline content to avoid shell injection and using --force-with-lease to prevent accidental overwrites during pushes.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository files to automate PR creation, which is an inherent risk for agents interacting with third-party codebases.
Ingestion points: The skill reads PR templates, contributor guides, and existing PR metadata from the target repository to resolve language and structure (documented in SKILL.md and references/pr-body-guidance.md).
Boundary markers: The skill utilizes temporary files via mktemp and the --body-file argument to pass data to CLI tools, creating a clear boundary between data content and command execution (documented in references/publish-workflow.md).
Capability inventory: The skill is capable of repository mutation (push, rebase), pull request lifecycle management (create, edit, merge), and temporary file handling (documented in references/publish-workflow.md).
Sanitization: The skill relies on strict workflow constraints and the use of specialized CLI flags for data handling rather than explicit programmatic sanitization of the ingested template content.

draft-pr