agent-inbox
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from the
leverage inbox --unreadcommand, which may contain messages from other agents or users. - Ingestion points: Output from
leverage inbox --unread(referenced in SKILL.md). - Boundary markers: No explicit delimiters or 'ignore instructions' prompts are used when processing inbox data.
- Capability inventory: The agent is instructed to read and process these messages to update the user on pending tasks.
- Sanitization: No sanitization or validation of the inbox content is performed before processing.
- [COMMAND_EXECUTION]: Executes local CLI commands (
leverage whoami,leverage inbox,leverage approvals) to manage agent state and fetch user information. - [EXTERNAL_DOWNLOADS]: Recommends installing related skills from the vendor's official repository using
npx skills add leverage-computer/skills.
Audit Metadata