journal-adapt

Pass

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes external CLI tools to perform PDF-to-Markdown conversion. Specifically, it uses python3 -m mineru.cli.pdf_to_md to process research papers and manuscripts provided by the user. While this is a functional requirement for the skill's purpose, it involves running shell commands on local files.
  • [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection. It ingests untrusted data from a corpus of research papers and user-provided exemplars to generate a 'dynamic_writing_skill.md' file, which is then loaded as the authoritative instruction set for the second phase of the agent's execution. A malicious document in the input corpus could contain hidden instructions designed to manipulate the agent's behavior during the revision phase.
  • Ingestion points: Step 2 (Corpus conversion) and Step 3 (Style Card extraction) ingest content from external PDF/Markdown files.
  • Boundary markers: The skill lacks robust technical delimiters to separate untrusted document content from the agent's internal reasoning, relying instead on high-level natural language instructions to 'describe only structure'.
  • Capability inventory: The agent has the capability to write files to the local file system and execute shell commands (mineru).
  • Sanitization: There is no evidence of automated sanitization or filtering of the text extracted from the external papers before it is processed by the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
May 16, 2026, 04:24 PM
Security Audit — agent-trust-hub — journal-adapt