The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the GitHub CLI (gh) and git to function, executing these tools through the Python subprocess module. This is the primary method for gathering PR data and performing the closing actions.\n
Evidence: Found in scripts/collect_prs.py and scripts/deplib/ghcli.py.\n- [INDIRECT_PROMPT_INJECTION]: The agent processes untrusted data from external sources (pull request titles, bodies, and diffs from GitHub). This content is presented to the agent, which could potentially contain instructions aimed at influencing the agent's logic or decision-making process.\n
Ingestion points: scripts/collect_prs.py fetches PR metadata and diffs using the GitHub API/CLI.\n
Boundary markers: The skill instructions do not specify explicit delimiters or isolation markers when presenting PR data to the agent in Step 3.\n
Capability inventory: The skill has the capability to close pull requests and delete branches via the gh pr close command.\n
Sanitization: While the scripts parse the diffs to identify manifest changes, other fields like titles are used directly in the overview and proposals presented to the user.\n- [DYNAMIC_EXECUTION]: The library includes a mechanism to override the gh command via the DEP_PR_CLEANUP_GH environment variable. If set to a path ending in .py, the skill will execute that file using the current Python interpreter instead of the standard CLI.\n
Evidence: Logic located in scripts/deplib/ghcli.py.

dep-pr-cleanup