The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository environment (such as PR descriptions, commit messages, and documentation files) and interpolates it into the agent's context to generate pull request content.
Ingestion points: Metadata and history are read in SKILL.md (Sections 1 and 2) and through lookup commands in references/github.md and references/gitlab.md.
Boundary markers: The skill identifies and filters bot-generated content using specific HTML comment markers, though it lacks general delimiters for other user-contributed content.
Capability inventory: The skill can write to the local filesystem (e.g., .git/PR_BODY.md), execute a bundled Python script for linting, and perform authenticated API operations via gh or glab CLIs to create or update PRs.
Sanitization: The skill uses scripts/lint_body.py to enforce style and structure. Crucially, it implements a manual confirmation step (§5. Confirm inline) that requires the user to approve the drafted title and body before any changes are pushed to the remote host.
[COMMAND_EXECUTION]: The skill performs shell-based operations using git, gh (GitHub CLI), glab (GitLab CLI), and python3 to manage repository state and validate drafted content. These commands are well-scoped to the skill's primary purpose and utilize secure practices, such as passing content via files (--body-file) to prevent shell injection vulnerabilities.

create-pr