aris-paper-figure

Pass

Audited by Gen Agent Trust Hub on May 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses a Bash loop in Step 5 to run dynamically generated Python scripts (python "$script"). This is a core part of the workflow but presents a surface for executing unintended code if the generation process is compromised.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data to drive its logic without sufficient boundaries.
  • Ingestion points: The skill reads PAPER_PLAN.md and experiment data files (JSON/CSV) to determine figure types, descriptions, and captions.
  • Boundary markers: No delimiters or safety instructions are defined to prevent the agent from following instructions embedded within these data files.
  • Capability inventory: The agent possesses powerful capabilities including Bash, Write, and the mcp__codex__codex tool.
  • Sanitization: There is no evidence of sanitization or validation of the input data before it is used to construct Python scripts or prompts for the external reviewer model.- [DATA_EXFILTRATION]: Local metadata, such as figure descriptions and captions extracted from the user's project, is transmitted to an external service via the mcp__codex__codex tool for quality review in Step 7.
Audit Metadata
Risk Level
SAFE
Analyzed
May 6, 2026, 01:42 AM
Security Audit — agent-trust-hub — aris-paper-figure