The live, agent-readable frontier of AI mathematics. Provides read-only access to normalized problems, AI attempts, partial progress, solution claims, verification evidence, method families, and replay-ready opportunities.
VibeMath demonstrates solid definition quality with complete input schemas, clear descriptions, and well-structured parameters. All 6 tools have explicit JSON Schema definitions with type constraints and enums. Descriptions are concise (50-150 chars) and action-oriented. However, output schemas are not documented in the tool definitions themselves, only inferred from implementation. Parameter descriptions are present but could be more prescriptive about use cases and dependencies. No tool annotations (readOnlyHint, destructiveHint) despite all tools being read-only. Error handling is minimal, no recovery guidance or actionable error messages visible in tool definitions.
Export a stable candidate-only handoff for Low-Hanging Fruit.
Get one problem with attempts, events, evidence, verification, and replay status.
List transparent replay-triage candidates; this is not a success probability.
Return source freshness, license, schema, and coverage limits.
List recent source-reported AI mathematics events from the local snapshot.
Search normalized problems without claiming complete mathematical coverage.
Output schemas not documented in tool definitions. LLMs cannot plan downstream calls or extract required fields without seeing what each tool returns.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Agents cannot distinguish safe tools from destructive ones without explicit hints.
Parameter descriptions lack prescriptive guidance on use cases and dependencies. E.g., 'verification' filter in latest_ai_math_events does not explain when to use it or how it interacts with 'status'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Tool 'export_low_hanging_fruit_handoff' has ambiguous naming. 'export' suggests data export, but the description indicates it builds a handoff object. Consider 'build_lhf_handoff' or 'prepare_handoff_for_lhf'.
No error handling guidance in tool definitions. If a problem_id is invalid or a search returns no results, LLMs have no recovery path documented.