Sensorium
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| runsA | list recorded traces. One line per trace, oldest first: run id, exit, event count, the command; a refocus rerun names its original and its verdict. |
| infoA | summarize one trace. Recorder, language, the capabilities the trace declares, what was recorded and what was not, focus, the redaction stamp; read it before any other question on a run. |
| treeA | call-tree slice. Parentage derived from the event stream; |
| frameA | one activation in full. Arguments, return or raise, and the per-line locals timeline when the run was recorded with |
| grepA | search events by name or value. Every CALL, RETURN, RAISE, HANDLED or LINE event whose name or rendered value contains the pattern; |
| exceptionsA | raises, handles, swallows. Every raise classified swallowed, uncaught, re-raised, propagated or ambiguous, with the reason; SWALLOWED is claimed only when the recording proves it. |
| flowA | provenance of a value or an object. |
| watchA | predicate over captured state. A gdb-style watchpoint evaluated after the fact at every recorded site of one function; needs a |
| diffB | first causal divergence between two runs. Aligns the two causal streams and names the first event where they part, with drill-in commands; MATCH or DIVERGED, or REFUSED when no verdict exists. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Each of the nine tools owns a distinct slice of the trace-analysis workflow: run listing, run summary, call-tree navigation, frame detail, event search, exception classification, value/object provenance, predicate evaluation, and run comparison. Even close neighbors like tree/frame and grep/exceptions are cleanly separated by granularity and purpose. An agent should rarely misselect.
All tool names are concise lowercase single-word commands, so the naming style is predictable and visually consistent. The only minor deviation is a mix of nouns/verb forms like info, tree, frame vs. grep, watch, diff, but there is no mixed casing or chaotic variation. This is a small style inconsistency rather than a real usability problem.
Nine tools is an ideal scope for a specialized trace-analysis server: broad enough to cover the main investigative workflows, yet small enough that every command has a clear role. It is comfortably within the well-scoped range and does not feel over- or under-built.
The tool surface covers the full analytical lifecycle: discovering runs, reading summaries, exploring call trees and frames, searching events, classifying exceptions, tracing value/object provenance, evaluating predicates, and diffing two runs. The tools explicitly reference each other's outputs such as frame ids and event ids, enabling iterative investigation without dead ends. A raw event timeline is absent, but grep, tree, and exceptions cover that need effectively.