The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes multiple shell scripts to automate repository maintenance tasks. These include find-candidates.sh for PR discovery, collect-gates.sh for status check aggregation, check-coderabbit-threads.sh for review thread resolution via GitHub GraphQL, and parse-supersession.sh for identifying related PRs.
[DYNAMIC_EXECUTION]: Step 4 of the workflow executes a shell script directly from the repository's git history using bash <(git show origin/main:...). This is used to run project-specific gate checks from the canonical branch of the checkout.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issue descriptions, comments, and pull request bodies to determine merge eligibility.
Ingestion points: Content is fetched using gh issue view and gh pr view within SKILL.md and several helper scripts.
Boundary markers: The skill does not explicitly define delimiters for external text, but requires the evaluator to provide specific reasoning evidence, including file and line references, for every judgment to ensure traceability.
Capability inventory: The agent utilizes bash, python, git, and gh to read and evaluate repository state.
Sanitization: The skill relies on structured evaluation tiers (0, 1, 2) and mandatory human review of the generated scorecard before any merge action is recommended.

nemoclaw-maintainer-pr-comparator