art-director
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from concept packets and visual evidence cards, creating a surface for injection when this data is interpolated into prompts. 1. Ingestion points: Data enters the context via PIPELINE_CONTEXT and concept packets processed in SKILL.md and the prompt-templates/ directory. 2. Boundary markers: The skill uses bracketed placeholders (e.g., [APPROVED_STYLE_DNA]) in its templates but lacks explicit instructions to ignore instructions embedded within the interpolated data. 3. Capability inventory: The skill instructs the agent to execute shell and python scripts. 4. Sanitization: No sanitization, validation, or escaping of external inputs is documented.
- [COMMAND_EXECUTION]: SKILL.md and LITE.md direct the agent to execute specific shell commands using python3 and bash (e.g., 'scripts/art-direction/style-contract.py'). These scripts and the referenced '.forgewright/' configuration directory are not included in the skill package, which may result in execution failures or reliance on an unverified environment.
Audit Metadata