local-council-execution

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating untrusted data directly into subagent prompts and shell command arguments.
  • Ingestion points: User-provided questions, file content (via --file or auto-context), and the --roles command-line argument are ingested in SKILL.md (Step 1, Step 2) and agent-prompt-template.md.
  • Boundary markers: The subagent prompt template in agent-prompt-template.md does not employ delimiters (such as XML tags or explicit block markers) to isolate untrusted {QUESTION_WITH_CONTEXT} from the system instructions.
  • Capability inventory: The skill utilizes the bash tool to execute local plugin scripts, spawns multiple general-purpose subagents with background execution capabilities, and performs file system writes to a local cache directory (.claude/council-cache/).
  • Sanitization: No explicit validation, escaping, or sanitization is performed on the user-supplied role list or context strings before they are interpolated into shell commands (e.g., local_council_roles "<--roles value>") or subagent prompts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 11:41 PM
Security Audit — agent-trust-hub — local-council-execution