soia-dev-agent-cli-dispatch
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes Python's 'subprocess' module to orchestrate external AI command-line interfaces (CLIs) such as Codex, Claude Code, and Gemini CLI.
- Evidence: Located in 'scripts/run_matrix.py' and 'scripts/run_claude_prompt.py'.
- Context: This is the primary intended function of the skill. The instructions explicitly mitigate shell injection risks by mandating that prompts be passed via temporary files and stdin rather than as command-line arguments.
- [EXTERNAL_DOWNLOADS]: The documentation references a remote installation script for the Antigravity CLI.
- Evidence: 'https://antigravity.google/cli/install.sh' mentioned in 'references/antigravity-cli.md'.
- Context: The resource is hosted by a well-known service provider (Google) and is part of the standard administrative setup for the tool.
- [SAFE]: The skill exhibits an indirect prompt injection surface as it processes untrusted user data.
- Ingestion points: User prompts and task definitions in 'SKILL.md' and associated scripts.
- Boundary markers: The skill uses explicit boundary markers and 'Anti-Fake-Fix' validation gates to ensure that model outputs match expected results.
- Capability inventory: File system operations and CLI execution restricted by user-defined working directories.
- Sanitization: Employs stdin redirection and temporary file isolation to sanitize prompt interpolation into shell commands.
Audit Metadata