Loading ToolBench
Authoritative Multi-Agent Engineering Coordinator — a Tauri-based desktop application and daemon for coordinating multiple AI coding agents across Git repositories with task management, scope locking, verification, and merge queue orchestration.
AgentXFlow defines 22 tools with schemas and descriptions, but quality is inconsistent. Most tools have basic descriptions (50-150 chars) and structured input schemas, but lack depth in parameter documentation and output schema specification. Parameter descriptions are often minimal (e.g., 'Project ID' vs. actionable guidance). No visible output schemas documented. Error handling guidance is absent. Tool names follow verb_noun convention well (e.g., project_list, task_claim), but descriptions don't explain WHEN to use each tool or dependencies between them. Security-critical tools (task_cancel, scope_acquire) lack permission/audit documentation. Composition is reasonable, tools chain logically, but many parameters lack type constraints (enums, ranges) that would guide LLM selection.
Keep agent session and active scope leases alive.
Register an agent session with a canonical AI IDE platform and get an authoritative session token and agent_id.
Get the most recently prepared handoff, active project, and recommended next action for newly connected agents.
List blocker tasks that must finish before this task starts.
Claim the next batch of steps from an organized masterplan and allocate an isolated Git worktree.
Decompose raw masterplan into structured execution steps.
Output schemas not documented. Tools return complex objects (Task, MasterPlan, VerificationRun) but no response schema is visible in tool definitions. LLMs cannot plan downstream calls or extract required fields.
Parameter descriptions are minimal and lack actionable guidance. Most describe only the parameter name (e.g., 'Project ID', 'Task identifier') without format, constraints, or when to use. Descriptions should explain format, valid ranges, and dependencies.
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or actionable error messages. E.g., task_claim does not document what happens if the task is already claimed or if scope conflicts exist.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Get masterplan specification, current status, project identity, and decomposition instructions.
List all masterplans across all projects with status, step counts, and active handoffs.
Query plan progress stats, total steps, and step statuses.
Atomically save, parse, structure, and prepare a masterplan for agents.
Get architectural rules, contract hashes, and project metadata.
List all managed projects with their exact project IDs, repository paths, and target branches.
Atomically lock file glob patterns for exclusive write access.
Release held write locks back to the pool.
Cancel an active task, releasing all write scope leases, cleaning up worktrees, and reverting any masterplan steps back to PENDING.
Atomically claim a task and cut an isolated Git worktree on disk.
Mark a required task step completed with test or build evidence.
Get complete task details including steps, acceptance criteria, active scope leases, attempts, and verification results.
Get task details including prompt, status, and worktree path.
List all tasks in backlog or ready queue for a project.
Requeue a claimed chunk task back to masterplan pending steps, releasing held scope leases.
Submit task for coordinator verification and git mutation audit.
Security-critical tools lack permission/audit documentation. task_cancel (DESTRUCTIVE), scope_acquire (WRITE), and agent_register (WRITE) do not declare required permissions or audit trail expectations.
Missing enum constraints on categorical parameters. agent_register accepts 'name' and 'agent_type' as enums (good), but many other tools accept free-form strings (project_id, task_id, agent_id) without validation hints or format documentation.