autonomous-skill
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- Command Injection (MEDIUM): In Step 3, the variable $USER_TASK_DESCRIPTION is interpolated directly into the 'claude' command line. A malicious task description containing double quotes could be used to inject additional CLI arguments, potentially overriding safety settings or max turn limits.
- Indirect Prompt Injection (LOW): The skill processes task instructions from local Markdown files and interpolates them into agent prompts. 1. Ingestion points: .autonomous/task_list.md and progress.md. 2. Boundary markers: None; contents are directly concatenated. 3. Capability inventory: Filesystem access (cat, ls, find) and 'claude' CLI execution. 4. Sanitization: None.
- Dynamic Execution (LOW): The skill implements an autonomous 'while true' loop (Step 4) that executes commands repeatedly. This recursive behavior can lead to uncontrolled resource consumption if the termination conditions are subverted.
Audit Metadata