Loading ToolBench
The Sovereign Email & Identity Layer for AI Agents. Provides autonomous email management, OTP extraction, task orchestration, and agent fleet coordination via MCP protocol.
AgentBox Mail defines 23 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (dispatch_agent_task, claim_agent_task, get_account_by_address). Parameter descriptions exist but lack actionable constraints (e.g., 'status' enum values are documented but format/range guidance is sparse). Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent from tool definitions. Security concern: agent_token parameters appear in multiple tools (dispatch_agent_task, claim_agent_task, update_task_progress, complete_task, fail_task) but no guidance on token injection, validation, or scope declaration. Tools like triage_with_gemini and resolve_with_gemini call external LLM APIs but lack timeout, retry, or partial-failure guidance.
Atomically locks and claims an unassigned or targeted task for a worker agent.
Marks a task as completed with final status, test results, and completion notes.
Provisions a new email account (inbox) with optional display name.
Dispatches a structured work order/task from one agent (e.g. Jules) to another agent or capability pool with evidence and acceptance criteria.
Extracts OTP codes and action links from email subject and body text using regex and heuristics.
Marks a task as failed with error details and optional reassignment to another agent.
Retrieves account metadata (ID, display name, address, created_at) for a given email address.
Output schemas not documented. LLMs cannot predict response structure for any tool. Baseline requires 100% of A+ tools to have documented return types.
Agent token parameters (agent_token) exposed in tool inputs without scope declaration or permission-gate guidance. Tokens risk leaking into logs and traces. No guidance on token validation or injection prevention.
External LLM calls (triage_with_gemini, resolve_with_gemini) lack timeout, retry, partial-failure, or error recovery guidance. No documentation of what happens if Gemini API fails or times out.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | <=2025-11-25 | v2 |
Retrieves the cryptographic identity (public key, signing key, capabilities) for a registered agent.
Retrieves a single message by ID with full MIME content and extracted metadata.
Retrieves all messages for a given account with full metadata (from, to, subject, body, extracted OTP, links).
Retrieves the current server configuration including domain, SMTP settings, and feature flags.
Retrieves real-time server statistics including message counts, task counts, and system health.
Retrieves a task by ID with full metadata, status, audit trail, and acceptance criteria.
Retrieves the complete audit trail for a task including all status transitions, commits, PRs, and notes.
Lists all provisioned email accounts (inboxes) with their metadata.
Lists all tasks with optional filtering by status, priority, or assigned agent.
Parses raw MIME email content and extracts structured fields (from, to, subject, body_text, body_html).
Registers a new agent with cryptographic identity (public key, capabilities, scope).
Uses Google Gemini 3.5 to generate an intelligent response to an inbound email and optionally send it.
Sends an outbound email from a provisioned account to one or more recipients via configured SMTP relay.
Uses Google Gemini 3.5 to intelligently triage an inbound email, classify it, extract intent, and suggest routing.
Updates task lifecycle status ('running', 'testing', 'pr_opened') and logs intermediate git commits or test results to the audit trail.
Verifies a cryptographic signature from an agent using their registered public key.
List tools (list_accounts, list_tasks) lack pagination parameters (limit, offset, cursor) and result count documentation. Baseline requires pagination for list operations to prevent context window exhaustion.
Descriptions for utility tools (extract_otp, parse_email_mime, get_server_config, get_server_stats) are 50-70 chars, below baseline 194 avg. Lack context on WHEN to use them vs alternatives and what downstream tools expect.