Loading ToolBench
Local control plane for multi-agent coding: share MCP/tools/memory, remember via handoffs, prove with file receipts, improve by promoting skills only on real exit codes. Includes GraphTrail (code-graph MCP server) and MiseLedger (evidence-ledger MCP server).
Brigade's 12 tools have basic schemas and descriptions, but fall short of production quality. All tools have descriptions (10-80 chars, below the 194-char baseline), and input schemas are present with type definitions. However, descriptions are generic and lack actionable context for LLM selection. No output schemas are documented. Error handling is absent, tools do not guide recovery or categorize failures. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification (6 READ_ONLY, 6 WRITE). Parameter descriptions are minimal (e.g., 'The backup target alias' repeats the name without explaining what constitutes a valid alias). No pagination, no batch operations, no idempotency guidance. The tool set shows good separation of concerns (backup vs fleet, propose vs execute) but lacks the depth needed for reliable agent composition.
Execute an approved backup action
Get status of a backup operation
Get overview of all backup targets with their current status and findings
Propose a backup action for review
Check restore readiness for a backup target
Get status of a specific backup target
Execute an approved remediation action
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., does backup_overview return operation_ids for chaining to backup_operation_status?). Breaks tool composition.
Descriptions are 25 - 35 characters, well below the 194-char baseline. Generic text like 'Get status of a specific backup target' does not explain when to call this vs backup_overview, what constitutes a valid target_alias, or what the response contains. LLMs cannot reliably select the right tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Get overview of all fleet hosts with their current status and findings
Get status of a specific fleet host
Get incident bundle for a fleet scope
Propose a remediation action for a fleet incident
Get health status of a service on a fleet host
No tool annotations despite clear risk stratification. 6 tools are READ_ONLY, 6 are WRITE (propose/execute). Missing readOnlyHint and destructiveHint annotations prevent agents from reasoning about safety and idempotency. Agents cannot distinguish safe retries from irreversible operations.
No error handling guidance. Tools do not document what errors are possible, whether they are retryable, or what the LLM should do next. A failed backup_execute_action returns nothing actionable, the agent cannot self-correct or escalate.
Parameter descriptions are minimal and repeat the parameter name. 'The backup target alias' does not explain format, valid values, or how to obtain one. LLMs cannot validate input before calling the tool, leading to failed calls and wasted tokens.