The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the gh (GitHub CLI) tool to resolve pull request URLs and runs a local Node.js script at agents/pr-reviewer/scripts/fingerprint.mjs to generate unique keys for stored rules.
[DYNAMIC_EXECUTION]: The skill invokes a local Node.js script using parameters (finder, defect-class, and path) inferred from user-provided natural language input. This process involves assembling command-line arguments dynamically based on interpreted user intent.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from GitHub Pull Requests and incorporates user-provided instructions into its memory system, creating a potential surface for indirect injection.
Ingestion points: Pull Request references and natural language "facts" provided in the command arguments are ingested into the skill's workflow.
Boundary markers: The skill does not employ explicit delimiters or boundary markers when passing the PR reference or flags to the pr-reviewer sub-agent prompt.
Capability inventory: The skill utilizes sub-agent dispatch (Task), GitHub pull request access (gh or mcp__github__pull_request_read), and long-term memory persistence (mcp__lorekit__memory_write).
Sanitization: The skill relies on the sub-agent's own logic and the validation checks within the fingerprint.mjs script (checking against FINDERS and DEFECT_CLASSES) rather than performing pre-processing sanitization on the inputs.

pr-review