The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing the prkit CLI tool to interact with a local desktop application. It performs various read and write-like operations (drafting comments, generating guides) through this interface.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, untrusted data from GitHub pull requests, including diffs, comments, and threads. This data could contain malicious instructions. \n
Ingestion points: prkit pr read, prkit pr files, prkit inbox, prkit guide read, and prkit pr conversation in SKILL.md and references/commands.md. \n
Boundary markers: The skill instructions emphasize that writing to GitHub requires user interaction in the desktop app, acting as a manual gate. \n
Capability inventory: The skill can create local drafts (prkit comments add), initiate agent-based fixes (prkit fix send), and generate guides (prkit guide generate). \n
Sanitization: No explicit sanitization or filtering of the ingested pull request content is described in the provided instructions.
[PRIVILEGE_ESCALATION]: The instructions direct the agent to ask the user to allow network access or run the command outside the sandbox if loopback connections to the local app are blocked. This encourages the user to lower established security boundaries.

prkit-cli