Multi-agent orchestration platform for Claude Code, Codex CLI, and Gemini CLI
Multiagents provides 18 tools with consistent naming (verb_noun pattern) and reasonable descriptions (avg ~120 chars). All tools have input schemas with type definitions and required fields. However, output schemas are entirely undocumented, no tool specifies what fields are returned or their types. Parameter descriptions are present but often generic (e.g., 'Target peer ID' lacks context on format or resolution). Error handling is absent, no guidance on recovery, retryability, or actionable next steps. Tool composition is sound (single responsibility), but several tools lack depth in their descriptions regarding when/why to use them vs. alternatives. Security considerations (permissions, audit trails) are not documented.
Acquire an exclusive lock on a file for editing.
Approve another agent's work. This signals that their implementation meets quality standards. Only call when you are satisfied with the work.
Assign a role and description to a peer.
Manually poll for new messages from other agents.
See the full team status: every agent's role, connection state, task state, and what they are working on. Use this proactively to know who needs help, who is waiting for review, and who is blocked. Call this regularly.
Retrieve message history from the session log.
Get the session plan with all items, their IDs, statuses, and assignments. Call this to learn which plan items are assigned to you and what their IDs are, so you can update them with update_plan.
No output schemas documented for any tool. LLMs cannot infer return types, field names, or structure. Agents must guess what data is available after each call, increasing hallucination risk and forcing exploratory calls.
No error handling guidance. Tools lack descriptions of failure modes, retryability, or recovery steps. E.g., acquire_file does not explain what happens if the file is already locked, or how to handle conflicts.
Parameter descriptions lack actionable detail. 'Target peer ID' does not explain format, how to obtain it, or whether names are accepted. 'Brief summary' lacks length constraints or examples of good summaries.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
List other agent instances. Returns ID, Name, Type, Role, CWD, Summary, Last seen.
Query shared knowledge stored by teammates. Search by key prefix or category.
Release a file lock you hold.
Set or change a peer's display name.
Send a message to another agent instance by peer ID.
Set a 1-2 sentence summary of your current work (visible to peers).
Signal that your current task is complete and ready for review. Do NOT call this prematurely — only when your implementation is truly done. After calling this, stay active and wait for feedback.
Store shared knowledge visible to all teammates. Use for architectural decisions, discovered patterns, coding conventions, and project context. Knowledge persists across agent restarts. Query with query_knowledge.
Send review feedback to another agent. Set actionable=true if changes are needed (sends agent back to work), false for informational comments.
Update a plan item's status (pending, in_progress, done, blocked). Use this to mark your assigned plan items as you complete them. Call get_plan first to learn your item IDs.
View all active file locks and ownership assignments.
No pagination or result limits documented. get_history, check_team_status, and query_knowledge may return unbounded lists, risking context window exhaustion. No mention of limit defaults or maximum values.
No security or permission documentation. Tools like approve and submit_feedback lack clarity on who can call them, what permissions are required, or whether actions are audited. No mention of scope declarations or audit trails.