arc-ideabrowser-openclaw-flow
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external snapshots of IdeaBrowser pages, which could contain adversarial instructions intended to be interpreted by the agent during the generation of the scaffold spec or implementation plan.
- Ingestion points: The scripts
scripts/crawl_idea_subpages_playwright.shandscripts/crawl_idea_subpages.shfetch and read snapshots from external URLs. - Boundary markers:
SKILL.mdprovides explicit instructions to the agent not to write files or execute commands without user approval, though it does not explicitly warn the agent to ignore instructions embedded within the extracted data. - Capability inventory: The skill utilizes
playwright-cliandopenclawfor browsing and suggests shell commands for repository initialization, deployment, and git operations. - Sanitization: No explicit sanitization or filtering of the snapshot content is performed within the scripts before the agent processes the text.
Audit Metadata