research-prompt

Pass

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides instructions to execute research prompts via an external API endpoint (DeepAPI POST /v1/research/deep). This operation involves sending project context to a third-party service, which could lead to data exposure if sensitive information is included in the project files. The skill partially mitigates this by including a limitation requiring explicit user approval for remote access and environment interaction.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its reliance on external data sources for prompt generation.
  • Ingestion points: Project files and conversation history are used to populate the research context (SKILL.md).
  • Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions to isolate the ingested context from the agent's core instructions.
  • Capability inventory: The generated prompt is intended for execution via network-based researcher agents (DeepAPI).
  • Sanitization: No explicit sanitization, filtering, or validation of the ingested content is described in the prompt generation process.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 14, 2026, 10:58 AM
Security Audit — agent-trust-hub — research-prompt