latex-manuscript-format-converter

Warn

Audited by Gen Agent Trust Hub on Oct 24, 2026

Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTION
Full Analysis
  • [METADATA_POISONING]: The skill package includes a file titled eval_report_latex-manuscript-format-converter_result.json. This file is a fabricated security audit mimicking the output of an automated auditor (version 'skill-auditor@1.0'). It makes authoritative safety claims such as 'security: PASS', 'Production Ready', and 'deployable: true'. The inclusion of this file is a deceptive attempt to influence the agent's behavior and the system's security posture by bypassing legitimate inspection with a falsified 'safety certificate'.
  • [INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to ingest and process untrusted external data, creating an attack surface for indirect prompt injection.
  • Ingestion points: The skill (SKILL.md) explicitly instructs the agent to process 'manuscript draft', 'target journal instructions', 'conference template', and 'external Word/plain-text files' provided by the user.
  • Boundary markers: There are no instructions or technical rules defined in the skill (such as in references/hard-rules.md) to apply boundary markers or delimiters when presenting untrusted content to the model, nor are there instructions to ignore hidden directives within those files.
  • Capability inventory: The skill has the capability to generate and output structured LaTeX code based on these inputs. A malicious input could trick the agent into emitting malicious LaTeX commands or leaking parts of the system prompt under the guise of 'improving template alignment'.
  • Sanitization: The skill lacks any mechanism for sanitizing or filtering input text for embedded instructions before it enters the agent's context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 24, 2026, 05:04 PM