A generic multi-agent swarm engine with MCP support for Emergent Wisdom projects
Three tools with basic schemas and minimal descriptions. send_message and submit_product lack detail on side effects and prerequisites. read_history has no guidance on output format or pagination. No parameter validation hints, no error recovery guidance, and descriptions are generic (10-50 chars, well below the 194-char baseline). Tool names are verb-noun compliant but descriptions do not explain WHEN to use each tool or WHAT happens. No mention of idempotency, retryability, or state changes. Schemas present but sparse, no output schemas documented, no pagination support, no field descriptions beyond parameter names.
Read the full history of messages from a specific sender.
Send a message to a teammate.
Submit final artifact to end the mission.
Descriptions are too brief (10 - 50 chars) and lack context. 'Send a message to a teammate' does not explain when to use this vs direct agent communication, whether it's async, or what happens on failure.
No output schemas documented. LLMs cannot infer what send_message returns (success flag? message ID? timestamp?), forcing guesswork on downstream tool chaining.
read_history has no pagination support (limit, offset, cursor). If history is large, the tool will either truncate silently or blow the context window.
No error handling guidance. If send_message fails (recipient offline, permission denied), the LLM has no recovery path. Descriptions do not indicate retryability or side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Parameter descriptions are minimal. 'Name of the agent to send to' does not clarify whether agent names are case-sensitive, if aliases are accepted, or what happens if the name is ambiguous.