The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around executing shell commands via the bash tool to manage coding agents. It provides templates for running these agents in the background with pseudo-terminal (PTY) support and provides instructions for interactive process control (write, submit, kill).
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because its primary purpose is to have an AI agent process external, untrusted code from git repositories and pull requests.
Ingestion points: External repositories cloned via git clone or checked out via gh pr checkout (referenced in SKILL.md).
Boundary markers: The skill suggests using isolated working directories (workdir) and temporary folders (mktemp) to prevent the agent from accessing sensitive files, but it does not specify markers to help the LLM distinguish between its instructions and the code being reviewed.
Capability inventory: The agent has access to bash for command execution and process tools for interaction, allowing it to perform significant actions on the host or in the sandbox.
Sanitization: There is no evidence of sanitization or filtering of the external code content before it is processed by the coding agents.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions to install an external package (@mariozechner/pi-coding-agent) from the npm registry and provides examples of cloning repositories from GitHub.

coding-agent