doc-workflow
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION] (MEDIUM): The skill uses the
!syntax inSKILL.mdto execute shell commands (test,find,cat,wc) at runtime for pre-fetching context. While the commands are currently hardcoded to specific paths, they represent active shell interaction. - [PROMPT_INJECTION] (LOW): Category 8: Indirect Prompt Injection surface detected. The skill automatically reads the first 30 lines of
llmdoc/index.mdinto the prompt context. - Ingestion points:
llmdoc/index.mdviacatcommand inSKILL.md. - Boundary markers: Absent; content is directly interpolated into the context.
- Capability inventory:
Read,Glob, andAskUserQuestiontools. - Sanitization: None; raw file content is piped to the agent.
- [EXTERNAL_DOWNLOADS] (SAFE): No external network requests or package installations were detected.
Audit Metadata