llm-council
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local script
.claude/scripts/code_graphwith variable file path arguments derived from the workspace or user queries. While these are local calls, they represent a surface for argument-based command injection if the underlying agent platform lacks robust sanitization.- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user questions and project documentation into prompts for advisors and reviewers, creating a surface for indirect attacks. - Ingestion points: User input and local project files (e.g.,
CLAUDE.md,AGENTS.md,docs/). - Boundary markers: Utilizes triple-dash delimiters (
---) and explicit block labeling to isolate data from instructions. - Capability inventory: Reading workspace files, writing local reports to
tmp/reports/, executing local scripts via subprocess, and spawning subagents. - Sanitization: No explicit logic for sanitizing or escaping the data interpolated into LLM prompts is provided.- [DYNAMIC_EXECUTION]: The skill uses
spawn_agentto create subagents at runtime and dynamically generates command-line calls for local analysis tools based on the current context.
Audit Metadata