saul-goodman-perspective
Pass
Audited by Gen Agent Trust Hub on May 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several scripts in the 'scripts/' directory, such as 'merge_research.py', 'quality_check.py', and 'download_subtitles.sh'. These tools allow the agent to execute shell commands and Python code within the host environment for research maintenance and subtitle processing.\n- [EXTERNAL_DOWNLOADS]: The 'scripts/download_subtitles.sh' script utilizes 'yt-dlp' to fetch media assets from external YouTube URLs. Additionally, the 'SKILL.md' file defines a workflow ('Step 2: Saul 式研究') that explicitly directs the agent to perform web searches to retrieve and summarize plot information from well-known entertainment and news websites.\n- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it combines external data ingestion with local capabilities.\n
- Ingestion points: External content is ingested via 'WebSearch' results as part of the core fact-checking protocol in 'SKILL.md'.\n
- Boundary markers: The skill's instructions do not specify any delimiters or safety markers to isolate search results from the agent's primary instructions.\n
- Capability inventory: The agent has the ability to execute local scripts and standard system tools.\n
- Sanitization: There are no provided instructions for sanitizing or filtering external content, which could allow maliciously crafted web data to influence the agent's persona-driven actions.
Audit Metadata