skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/n8n-photo-report/Gen Agent Trust Hub
n8n-photo-report
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [SAFE]: The skill implementation follows secure coding practices for its intended purpose. No malicious commands, obfuscation, or hardcoded credentials were found.
- [PROMPT_INJECTION]: The skill processes construction images through an AI vision model to generate summaries and safety alerts, which introduces a surface for indirect prompt injection.
- Ingestion points: Data enters the workflow via the 'Photo Webhook' and the 'Watch Dropbox Folder' node as seen in SKILL.md.
- Boundary markers: The AI prompt in the 'Analyze with Claude Vision' node does not currently employ delimiters or specific instructions to ignore text instructions found within images.
- Capability inventory: The skill can write to Airtable databases and send emails to recipients defined in environment variables.
- Sanitization: Output from the AI is parsed as JSON but not specifically sanitized for malicious payloads before inclusion in email reports.
- [EXTERNAL_DOWNLOADS]: The workflow performs network operations to interact with well-known services (Anthropic and Airtable). These connections are necessary for the skill's functionality and use official API endpoints.
- [COMMAND_EXECUTION]: The provided Python helper script in SKILL.md reads local files for upload to the workflow. This aligns with the 'filesystem' permission requested in claw.json.
Audit Metadata