lead
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to use a CLI tool ('bb') to manage project state and spawn sub-agents.
- Evidence: Frequent use of commands such as
bb status,bb thread spawn, andbb thread context. - Evidence: The instructions specifically direct the agent to spawn workers with the
--permission-mode autoflag, which grants the sub-agents autonomy to execute commands without further user confirmation. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data which could contain malicious instructions.
- Ingestion points: User-provided goals, project-specific playbooks (e.g., 'feedback-leader-playbook'), and repository evidence including code and specifications.
- Boundary markers: The skill suggests using
--prompt-fileor shell heredocs (EOF) to pass data to sub-agents, though it does not mandate specific delimiters for untrusted input within those files. - Capability inventory: The 'leader' agent has the ability to spawn multiple sub-agents (
bb thread spawn) and write adjudication documents directly to the repository. - Sanitization: The instructions include a specific security warning to the agent to avoid shell injection when using double quotes by being mindful of backticks.
Audit Metadata