The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-controlled data such as PR titles, descriptions, and commit messages to automate workflow decisions, including classifying review depth and selecting merge strategies.
Ingestion points: Data is ingested via gh pr view and git rev-list as defined in SKILL.md (Part R and Part D).
Boundary markers: The skill does not implement explicit delimiters or boundary instructions to prevent the agent from following commands embedded within pull request metadata.
Capability inventory: The skill possesses capabilities to execute shell commands (git, gh, glab), perform force pushes to personal branches, and run local Python scripts.
Sanitization: No explicit sanitization of ingested text is performed before it is used to influence decision-making logic.
[DYNAMIC_EXECUTION]: In Part C, the skill executes a Python script (analyze_coverage.py) from a directory associated with the test-coverage-analyzer skill using a computed path.
[COMMAND_EXECUTION]: The skill frequently invokes CLI tools like git, gh, and glab to perform repository management operations, including git push --force-with-lease during the rebase phase.
[EXTERNAL_DOWNLOADS]: The documentation instructs users to install dependencies from the vendor's repository using npx skills add FuDesign2008/open-skills, which facilitates the download and execution of remote code.

merge-discipline