Loading ToolBench
Coordinates human-authorized environmental fieldwork. Browser stations control sensors and physical action; MCP agents may design missions, compare structured evidence, and return visible findings.
Sensorium defines 12 tools with consistent verb-noun naming (inspect_, read_, compare_, list_, score_, create_, validate_, request_, wait_for_). All tools have descriptions (avg 95 chars, within 10-1024 range). Input schemas are present with type definitions and descriptions. However, output schemas are not documented in the source code, LLMs cannot predict return structures. Parameter descriptions are adequate but lack format constraints (e.g., bridgeCode length, expeditionCode format). Error handling is minimal; no recovery guidance or actionable error messages visible. Tool composition is sound (single responsibility), but some tools like wait_for_station_update lack timeout/pagination guidance.
Rank the current Sensorium samples using the same habitat scoring model as the browser instrument.
Rank all live stations in an expedition by their latest sample scores using the expedition's goal profile.
Create a six-hour multi-station investigation that browser participants can join with an expedition code.
Explain the remote Sensorium collaboration model and the evidence available to MCP clients.
List the live browser stations participating in a Sensorium expedition and their latest sample metadata.
List the goal-specific scoring profiles Sensorium can apply to the same physical evidence.
Output schemas not documented. LLMs cannot predict return structures for any tool, forcing them to guess field names and types for downstream chaining.
Parameter format constraints missing. bridgeCode, expeditionCode, sampleId lack length/pattern documentation. LLMs may pass invalid values without guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
Read the current investigation and structured environmental samples intentionally shared from a Sensorium browser.
Send a visible eight-second recapture mission to one browser bridge when its evidence is not reliable enough for comparison.
Send one clear physical measurement mission to every live station, or to one station. The request becomes visible in each browser and still requires human action.
Re-score every sample in one bridge for focus, sleep, reading, video calls, or recording.
Inspect confidence, duration, reading count, signal variability, and motion validation for one bridged sample. Uses the latest sample when no ID is supplied.
Block until a Sensorium expedition receives a new update (sample capture, station join, agent finding, etc.), then return it. Useful for agents that want to react to human fieldwork in real time.
No error handling guidance. Tools lack recovery hints (e.g., 'Bridge not found. Check bridgeCode format or create a new expedition.'). Agents cannot self-correct on failures.
wait_for_station_update lacks timeout/pagination guidance. Blocking indefinitely risks agent hangs; no max_wait_seconds or result_limit documented.
Destructive tools (create_expedition, request_bridge_recapture, request_station_observation) lack confirmation/dry-run patterns. Agents cannot preview side effects before committing.