hung-yi-lee
Pass
Audited by Gen Agent Trust Hub on Jun 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run a local utility script,
scripts/hungyi_kb.py, for searching transcripts and navigating the knowledge graph. Static analysis of this script identifies usage ofsubprocess.run(), which executes shell commands. While used for legitimate functions like metadata synchronization, this represents a capability for arbitrary command execution if input is not strictly validated. - [PROMPT_INJECTION]: A transcript file (
raw/youtube/transcripts/dWQVY_h0YXU.md) was flagged for containing instruction-override patterns (e.g., "Ignore previous instructions"). Although these are likely part of the spoken lecture content rather than malicious intent, they present an indirect prompt injection surface as the agent is directed to ground its answers in these files. - [EXTERNAL_DOWNLOADS]: The
requirements.txtfile specifies external dependencies (youtube-transcript-api,networkx,python-louvain). These are standard libraries used for the skill's stated purpose of fetching YouTube data and processing graph structures.
Audit Metadata