The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided input through the $ARGUMENTS variable and <fact> tokens, which are directly interpolated into sub-agent prompts and a persistent memory store.
Ingestion points: $ARGUMENTS in SKILL.md is used to capture PR references, flags, and instructions for the remember operation.
Boundary markers: No explicit delimiters or 'ignore' instructions are used when passing user-provided flags to the pr-reviewer sub-agent.
Capability inventory: The skill utilizes gh CLI, GitHub MCP tools for reading PRs, sub-agent dispatch tools (Task/Agent), node for script execution, and mcp__lorekit__memory_write for persistent storage.
Sanitization: The skill performs basic structural validation for the remember command, requiring specific fields (finder, class, path) before writing to memory, and explicitly refuses to suppress certain categories of findings.
[COMMAND_EXECUTION]: The skill executes local commands to interact with the environment and process data.
Evidence: Uses gh pr view to retrieve GitHub PR metadata and executes a local script via node agents/pr-reviewer/scripts/fingerprint.mjs to calculate rule fingerprints.

pr-review