Loading ToolBench
A gateway and toolkit for running AI agents with MCP (Model Context Protocol) support, featuring multiple adapters (Slack, Buzz, Console), job scheduling, team knowledge integration via ox, and credential management.
SageOx Agent Toolkit defines 9 tools with complete input schemas and descriptions. Naming follows verb_noun convention (brain_list, brain_read, brain_write, job_run, team_search). Descriptions are present and contextual (avg ~100 chars), explaining WHAT and WHEN. However, parameter descriptions are sparse or missing for several tools (e.g., brain_consolidate has no parameters documented; team_sessions/team_recent lack detail on 'repo' semantics). Output schemas are not documented in the source. Error handling guidance is absent, tools do not indicate what failures mean or how to recover. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (READ_ONLY vs WRITE). Composition is sound: tools are single-purpose and chainable (e.g., brain_read → brain_write). Security: no credentials in parameters, but no audit trail or permission gate documentation visible.
Report near-duplicate facts in this brain for review. Report only: it never deletes or rewrites notes.
List what is in your brain — file names only, no contents. The cheap first look before reading.
Read your brain. With a query, returns matching notes; without one, everything. Says so explicitly when nothing matches.
Append one fact to your brain with its source. Write before the turn ends — anything not written down is lost when the turn stops.
Start a job with optional typed parameters. The job must declare onRequest trigger and the parameters must match the job's declaration.
List recent activity (murmurs and sessions) in a repository within a time window.
Output schemas not documented. Tools return results but LLMs cannot plan downstream calls without knowing field names and types. brain_read, team_search, team_sessions, team_recent lack response structure documentation.
Tool annotations missing. brain_write and job_run are destructive (WRITE risk) but lack destructiveHint. brain_list, brain_read, team_* are read-only but lack readOnlyHint. Agents cannot distinguish safe from unsafe operations.
Parameter descriptions incomplete or missing. brain_consolidate has empty properties {}; team_sessions and team_recent lack detail on 'repo' semantics (is it a name, ID, or URL?). job_run 'params' description is generic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Search the team's knowledge base via ox. Returns passages with scores and sources.
List sessions in a configured repository from the team ledger.
Check the status of team ledger sync and code index readiness.
No error handling guidance. Tools do not document failure modes, recovery steps, or actionable error messages. E.g., brain_read says 'nothing matches' but does not guide what to do next.
Pagination limits not enforced in descriptions. team_search accepts 'limit' (max 20) and team_sessions/team_recent accept 'limit' (max 20) but descriptions do not state defaults or explain why limits exist.