deep-investigate
Pass
Audited by Gen Agent Trust Hub on Jun 21, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill logic is focused entirely on information gathering and documentation workflows. It uses standard reconnaissance tools to assist the user in exploring complex topics and does not contain any suspicious or hidden commands.
- [EXTERNAL_DOWNLOADS]: The skill uses web search and fetch capabilities to research external topics. These operations are essential to its stated purpose and target content relevant to the user's investigation requests.
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from external websites and codebase files, which is then passed to sub-agents. However, the risk is mitigated by the use of structured prompt templates.
- Ingestion points: Phase 1 (Scout) and Phase 3 (Fan out) gather data from
WebSearch,WebFetch, and local file reads. - Boundary markers: The sub-agent prompt utilizes explicit section headers (e.g.,
TOPIC:,SCOPE:,STARTING ANCHORS:) to differentiate instructions from processed data. - Capability inventory: The skill uses the
Agenttool to spawn sub-agents and performs local file writing to thehistory/directory. - Sanitization: No explicit sanitization or filtering of external content is performed before it is interpolated into the sub-agent prompt.
Audit Metadata