The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from GitHub issues and comments, creating a surface for indirect prompt injection.
Ingestion points: Data is retrieved via gh issue view <n> --comments, which includes text from external users.
Boundary markers: There are no instructions for using delimiters or boundary markers when interpolating issue content into the agent's context.
Capability inventory: The agent possesses significant capabilities, including shell command execution, running test suites, and modifying repository metadata (labels, comments, issues).
Sanitization: The instructions do not specify any sanitization or validation of the retrieved issue content before processing.
[COMMAND_EXECUTION]: The skill relies on extensive shell command execution to perform its auditing functions.
It uses utilities such as grep, find, ls, and sed to examine the filesystem and codebase.
It executes build and test commands including pnpm test, bun test, and npx node-gyp rebuild to verify implementation against documentation.
It invokes other local agent skills and scripts (e.g., claim.ts, link.ts, /agentloop:verification) to manage concurrency and verification logic.
[EXTERNAL_DOWNLOADS]: The 'Research' task type explicitly instructs the agent to clone external repositories for analysis.
The skill performs a shallow clone of external projects to a local 'scratchpad' directory to examine their architecture and source code. While limited to a temporary directory, this involves interacting with external infrastructure.

issue-review