The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A sensitive Google Gemini API key is hardcoded directly in the source code.\n
Evidence: index.js contains this.apiKey = "AIzaSyCKWmPmAkZWvI2KiblawWPUESyCp9dEjk0";.\n- [COMMAND_EXECUTION]: The skill documentation explicitly instructs users to use dangerous operational flags and configurations.\n
Evidence: SKILL.md recommends the --yolo flag for Codex, which is described as having "NO sandbox, NO approvals".\n
Evidence: The skill requires pty:true for interactive shell access, which increases the impact of potential command execution attacks.\n- [EXTERNAL_DOWNLOADS]: The skill documentation prompts the user to download and install global packages from external sources.\n
Evidence: SKILL.md suggests running npm install -g @mariozechner/pi-coding-agent.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through the processing of untrusted external content such as Pull Requests.\n
Ingestion points: SKILL.md describes cloning and checking out PRs for review.\n
Boundary markers: Absent; no markers are used to isolate untrusted code from agent instructions.\n
Capability inventory: The agent has access to the bash tool with pty, background, and elevated permissions.\n
Sanitization: Absent; no validation or sanitization is performed on the untrusted codebase content before it is processed by sub-agents.

coding-agent