broadcast
Pass
Audited by Gen Agent Trust Hub on Mar 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill performs extensive shell operations including git worktree management, build processes, and language-specific test suites (e.g., npm test, pytest, cargo test). It automatically scans broad directory paths such as $HOME/personal, $HOME/work, and $HOME/projects for repository discovery. It also generates and applies automated 'self-healing' code fixes.\n- [DATA_EXFILTRATION]: The skill's primary function includes pushing code changes to remote servers (git push) and creating pull requests via the GitHub CLI (gh pr create), which provides a mechanism for moving local code and configuration data to remote platforms.\n- [EXTERNAL_DOWNLOADS]: The skill synchronizes with external servers using git fetch and git pull commands to update local repository state.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the analysis and adaptation to local repository contents. 1. Ingestion points: Analysis of repository files, framework structures, and configuration formats in Phase 2. 2. Boundary markers: None implemented; the skill does not use delimiters to isolate repository-provided content. 3. Capability inventory: Extensive command execution, file system modification, and network operations via git/gh. 4. Sanitization: None present in the instructions.
Audit Metadata