research-proposal

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks because it processes untrusted user-supplied content (research proposals) through a series of autonomous sub-agent roles. This creates a surface where embedded malicious instructions could influence agent behavior.
  • Ingestion points: The proposal text is ingested from a user-defined path (<proposal_path>) or direct prose input as described in SKILL.md.
  • Boundary markers: The instruction sets for the sub-agents (roles/ScholarEval.md, roles/Judge.md, roles/Reviser.md) do not employ explicit delimiters or specialized formatting to isolate the processed proposal content from the agent's core instructions.
  • Capability inventory: The skill is capable of executing local Python scripts (the literature-search tool) which interact with academic databases, and it maintains a sandbox environment with read/write access to track iterations.
  • Sanitization: No validation or sanitization of the input proposal text is performed to mitigate the risk of embedded instructions affecting the LLM's logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 04:42 PM
Security Audit — agent-trust-hub — research-proposal