The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could contain malicious instructions.
Ingestion points: The skill fetches pull request metadata (title, body, commits) via gh pr view in Step 1 and analyzes code changes using git diff in Step 3.
Boundary markers: The instructions in Step 6 for drafting the updated PR do not include explicit delimiters or warnings to ignore instructions embedded in the ingested data.
Capability inventory: The skill can write files locally using apply_patch and modify remote GitHub repositories via gh api and gh pr edit in Step 8. It also has the capability to attach files from absolute paths to the PR.
Sanitization: While the skill uses file-based inputs in shell commands to prevent shell injection, it lacks prompt-level sanitization or filtering for the data ingested from the PR or git history.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with git and the GitHub CLI.
Evidence: Executes gh pr view, git fetch, git diff, gh api, and gh pr edit across various steps.
Context: These commands are necessary for the skill's functionality and are used with specific arguments that limit their scope to PR management.

update-pr