The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository logs and existing pull request content.
Ingestion points: The agent reads commit history via git log and git diff, and current pull request metadata via gh pr view as described in SKILL.md.
Boundary markers: The instructions define  markers for content preservation; these delimiters are present but do not serve as safety boundaries for instruction isolation in SKILL.md.
Capability inventory: The skill utilizes git and the GitHub CLI (gh) to read from and write to the repository environment, and performs file-write operations when saving the PR body to a temporary file, as specified in SKILL.md.
Sanitization: The skill includes a specific mitigation instruction to use the --body-file flag when updating pull requests, which prevents the shell from interpolating potentially malicious characters found in the processed data (SKILL.md).
[COMMAND_EXECUTION]: The skill relies on local shell commands to interact with the version control system and project management tools.
Evidence: The agent is instructed to execute git log, git diff, and gh pr view to gather context for the pull request body generation in SKILL.md.

pr