The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends the installation of the Node.js package '@mariozechner/pi-coding-agent'. This package is maintained by an individual contributor and does not originate from a trusted organization or well-known service.
[COMMAND_EXECUTION]: Explicit instructions are provided to execute powerful external tools while intentionally disabling their security and permission guardrails. Specifically, the skill directs the agent to use '--permission-mode bypassPermissions' with Claude Code and '--full-auto' or '--yolo' with Codex to suppress user-in-the-loop approvals for file system modifications.
[REMOTE_CODE_EXECUTION]: By automating external agents that generate and execute code locally while simultaneously disabling safety confirmations, the skill creates a path for unreviewed code execution on the host system.
[PROMPT_INJECTION]: The skill establishes a high-risk surface for indirect prompt injection by processing untrusted external content (cloned repositories and PR reviews) with tools that have had their safety filters disabled.
Ingestion points: External Git repositories and Pull Requests via 'git clone' and 'gh pr checkout'.
Boundary markers: Absent. The instructions do not include any delimiters or warnings to treat external code as untrusted.
Capability inventory: Full shell execution through multiple sub-agents (Claude, Codex, Pi, OpenCode) with automated approval flags.
Sanitization: Absent. There is no validation or sanitization of ingested code before it is passed to the execution tools.

coding-agent