deep-research
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from the web during its research phase, creating a potential surface for indirect prompt injection where malicious instructions could be embedded in search results.
- Ingestion points: The skill fetches content from multiple arbitrary web sources in 'Phase 3: FETCH AND EXTRACT FACTS' in SKILL.md.
- Boundary markers: The instructions mandate a structured fact extraction process (
F<id> | statement | source [n]...) and confidence scoring to isolate external data from instructions. - Capability inventory: The skill uses web search and fetch tools, and performs file system operations (directory creation and markdown file writing) in the
~/research-reportsand~/.claude/skills/deep-research/directories. - Sanitization: The skill implements a mandatory 'Adversarial Contradiction Pass' (Phase 4) specifically designed to seek out disconfirming evidence ('X debunked', 'X wrong'), which acts as a robust architectural control against biased or malicious external content.
Audit Metadata