Loading ToolBench
MCP server for authoring and validating Actlane packs with source contract inspection, validation, planning, and generation preview capabilities
Critical failures across all dimensions. All 11 tools expose input schemas with only `{"type":"object","additionalProperties":true}`, no parameter definitions, types, or constraints visible. Descriptions are present but generic (10-80 chars), lacking WHAT/WHEN/WHY guidance. No output schemas documented. No parameter descriptions. No error recovery guidance. Tool names use underscores but lack clear verb-noun structure (e.g., 'actlane_pack_inspect' vs 'inspect_pack'). No evidence of idempotency, pagination, or composition patterns. This server fails baseline quality checks for production agent use.
Classify a user task against Actlane runtime profiles and return advisory mode, risk flags, candidate capabilities, and required evidence.
Return compact evidence captured during this Actlane MCP session by id or latest marker.
Return a compact Actlane capability view derived from source contracts without mutation or execution.
Apply explicitly confirmed source contract files from a prior plan.
Explain pack load or validation errors in human-readable form.
Preview generated target files by delegating to the Actlane generator in a temporary directory.
All 11 tools expose empty input schemas: `{"type":"object","additionalProperties":true}`. No parameter names, types, descriptions, or constraints are defined. LLMs cannot infer what arguments each tool accepts or what they control.
No output schemas documented. Tools return JSON (e.g., inspectResult, validationInfo) but LLMs have no schema to understand response structure, field types, or what to extract for downstream calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Inspect an Actlane pack source tree without modifying files.
Plan source contract files for a new pack or capability without writing files.
Validate an Actlane pack and return structured diagnostics.
Prepare a final read-only delivery summary from the latest Actlane run and compact evidence.
Evaluate and prepare a guarded capability run through Actlane policy and MCPBinding-derived downstream plan.
Descriptions are generic and under 80 characters. E.g., 'Inspect an Actlane pack source tree without modifying files.' lacks WHEN to use, prerequisites, or what fields the response contains. No guidance on dependencies between tools.
No error recovery guidance. Tools return errors (e.g., 'pack load error') but do not tell LLMs what to do next: retry? Call a different tool? Ask the user? Agents cannot self-correct.
Tool names do not follow verb_noun convention. 'actlane_pack_inspect' is awkward; 'inspect_pack' is clearer. Prefix 'actlane_' is redundant given server context. Names like 'actlane_classify' and 'actlane_run_capability' lack clear action verbs.