setup-agent-config
Warn
Audited by Gen Agent Trust Hub on Apr 3, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The REFERENCE.md file refers to an external script located at '../shared/user-prompt-context.sh' intended for use as a UserPromptSubmit hook. This script is not provided in the skill files, which prevents verification of its behavior and implementation.
- [COMMAND_EXECUTION]: The skill provides bash scripts (llm-test-flags.sh and llm-truncate.sh) that are executed as platform hooks. These scripts have the capability to rewrite user-initiated tool commands in real-time and to suppress or replace tool output before it is returned to the user or agent.
- [PROMPT_INJECTION]: The documentation describes a mechanism to inject project metadata, including git state, file lists, and 'session violations', into every prompt via the UserPromptSubmit hook. This represents an indirect prompt injection surface. Evidence Chain: 1. Ingestion points: UserPromptSubmit hook (referenced in REFERENCE.md); 2. Boundary markers: Absent or not documented; 3. Capability inventory: Bash execution, tool input rewriting, and output suppression; 4. Sanitization: Absent or not documented in the provided logic.
Audit Metadata