tao-launch-workflow
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill implements a robust pre-launch gate for machine learning workflows, ensuring that all necessary credentials and compute configurations are verified before any side-effecting operations occur.
- [INDIRECT_PROMPT_INJECTION]: The skill serves as an intake gate for user-provided data such as dataset paths and compute configurations. These inputs are used to generate shell commands and configuration specs. Ingestion points: User inputs for dataset roots, spec keys, and platform credentials in SKILL.md and references/intake-prompts.md. Boundary markers: Uses specific prompt templates to delimit user inputs. Capability inventory: Executes shell scripts via the Bash tool to launch jobs and perform preflight checks. Sanitization: Employs redact_secrets.py to filter secrets from command strings and instructs the agent to verify variable presence rather than content.
- [COMMAND_EXECUTION]: The skill executes various local utility scripts and platform-specific CLIs (e.g., docker, sbatch, aws, kubectl) based on the selected execution environment. These are documented as part of the workflow orchestration.
- [EXTERNAL_DOWNLOADS]: Mentions downloading and installing missing CLI tools (e.g., aws) and Python dependencies at runtime. This is performed with user approval and targets well-known services.
Audit Metadata