The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell command execution using the GitHub CLI (gh) and uv to interact with pull requests and run helper scripts. Arguments are handled through safe mechanisms like json.dumps and integer casting in argparse to minimize command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill reads and acts upon PR comments and CI failure logs, which serve as external ingestion points for data that could contain malicious instructions.
Ingestion points: scripts/fetch_pr_feedback.py (line 297, 321) and scripts/fetch_pr_checks.py (line 184) ingest PR conversation and CI log data.
Boundary markers: There are no explicit instructions or delimiters in the skill workflow to prevent the agent from interpreting embedded instructions within the fetched comments or logs as authoritative.
Capability inventory: The skill provides the agent with the capability to edit files, commit changes, and push to the repository, which could be misdirected by a successful injection.
Sanitization: No sanitization or filtering logic is present to identify or neutralize potential instruction injection within the processed text.

iterate-pr