academic-paper
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a comprehensive 'Knowledge Isolation' protocol in
references/anti_leakage_protocol.mdand a generator-evaluator contract gate inagents/draft_writer_agent.md. These features are designed to prevent hallucinated content and maintain strict adherence to user-provided research materials. - [COMMAND_EXECUTION]: The
formatter_agent.mdfile contains instructions for executing document processing commands using Pandoc and LaTeX (e.g.,pandoc,xelatex,tectonic). These are standard industry-standard tools for academic publishing and are used strictly for document conversion. - [PROMPT_INJECTION]: The skill includes an ingestion surface for materials from 'deep-research' in
agents/intake_agent.mdandagents/draft_writer_agent.md. It employs strict boundary markers (<phase4a_output>, etc.) and an Anti-Leakage Protocol to mitigate risks associated with untrusted data interpolation. - [SAFE]: Hidden metadata markers in
agents/draft_writer_agent.md, such as<!--ref:slug-->and<!--anchor:kind:value-->, are used for internal citation tracking. The skill includes specific sanitization rules, such as URL-encoding and hyphen escaping, to ensure these markers do not interfere with the agent's instructions.
Audit Metadata