autoresearch
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute the shell command 'open dashboard.html' on macOS to launch a generated file in the default web browser.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential attack surface for indirect prompt injection by ingesting and acting upon the content of external 'Target skills'.
- Ingestion points: The agent reads and analyzes the full content of a user-specified target SKILL.md file from the filesystem.
- Boundary markers: No specific delimiters or instructions are provided to the agent to treat the target skill's content as untrusted data or to ignore embedded instructions during analysis.
- Capability inventory: The agent has the capability to write files (SKILL.md, dashboard.html, results.tsv, results.json, changelog.md) and execute shell commands (open).
- Sanitization: There is no evidence of sanitization or filtering of the target skill's content before it is used to generate prompt mutations or recorded in logs.
- [EXTERNAL_DOWNLOADS]: The generated dashboard.html file is designed to load the Chart.js library from an external Content Delivery Network (CDN) when opened in a browser. This is documented neutrally as a feature of the live dashboard.
Audit Metadata