photocraft-cli-text

Pass

Audited by Gen Agent Trust Hub on Oct 13, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install necessary dependencies, such as the photocraft-harness, using npx skills add. It also manages the download and installation of official CLI and desktop binaries (DMG) for the PhotoCraft tool.
  • [COMMAND_EXECUTION]: The skill relies on executing Python scripts (e.g., bootstrap.py, cli.py, commands.py) located in a secondary harness skill directory to perform its primary tasks. These scripts are used for environment setup and for interacting with the PhotoCraft CLI gateway.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided materials including image assets, project files (.pcraft), and task instructions, presenting an indirect prompt injection surface. However, the skill provides comprehensive guidance on mitigating this risk:
  • Ingestion points: Processes external assets and project files within the SKILL_DIR/examples/ and references/ directories.
  • Boundary markers: Explicitly defines scenario contracts and boundary refusal instructions (in references/scenario-contract.md and examples/boundary-refusal.md) to prevent unauthorized modifications.
  • Capability inventory: Limited to project modification, layer manipulation, and image export through a controlled CLI harness.
  • Sanitization: Implements strict integrity checks using SHA-256 digests for all downloaded binaries, assets, and project manifests.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 13, 2026, 01:35 PM