skill-authoring-workflow

Pass

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute multiple local shell and Python scripts (e.g., add-a-skill.sh, test-a-skill.sh, check-skill-metadata.py) located in the scripts/ directory. These commands are intended for repository management and do not involve privileged operations or external downloads.
  • [PROMPT_INJECTION]: The workflow ingests external data in the form of workshop notes or research files from the research/ directory to generate new skills. This creates an indirect prompt injection surface where untrusted content could potentially influence the agent's output during the authoring process.
  • Ingestion points: Processes files provided by the user in the research/ directory.
  • Boundary markers: None explicitly defined in the workflow instructions.
  • Capability inventory: Executes local scripts for file creation, validation, and metadata checking.
  • Sanitization: The skill relies on the underlying scripts and human review phases to ensure content compliance.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 3, 2026, 06:33 AM