contributing-to-pi

Pass

Audited by Gen Agent Trust Hub on Mar 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for repository management and testing. Evidence includes git worktree add /tmp/pi-mono-main origin/main, npm install, npm run check, and gh issue view. These commands operate on the local filesystem within ~/Code/badlogic/pi-mono and temporary directories.
  • [EXTERNAL_DOWNLOADS]: The skill triggers external data retrieval via git fetch origin and npm install. These operations fetch source code and dependencies from GitHub and the NPM registry respectively.
  • [REMOTE_CODE_EXECUTION]: By running npm install, the skill executes lifecycle scripts and code from the badlogic/pi-mono repository's configuration. While necessary for development, this involves executing code from a remote, non-trusted-list source.
  • [PROMPT_INJECTION]: The instructions mandate a specific stylistic persona, requiring the agent to use a "human voice" and avoid "agent mush" or "clanker slop." These directives are intended to influence maintainer perception and potentially bypass AI-detection heuristics.
  • [DATA_EXFILTRATION]: The research notes outline a technical proposal for a tracking system using Typesense and Restate to ingest repository artifacts (issues, PRs, comments) into an external database. While a design proposal, it describes a workflow for systematic data extraction from the GitHub environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 18, 2026, 12:38 PM