Loading ToolBench
MCP server that navigates agents through graph-based workflows with stateless workflow navigation and management tools
11 tools with complete JSON schemas and descriptions, but significant quality gaps. Tool names lack action verbs (Init, Start, Current, Next are weak; should be initialize_workflow, begin_workflow, get_current_step, advance_workflow). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameter descriptions exist but lack constraint details (no ranges, patterns, or validation guidance). Output schemas are not documented, LLMs cannot predict return structure. Error handling is absent from descriptions. No security considerations visible (file path parameters are write-capable with no sanitization guidance). Composition is reasonable (single responsibility per tool), but parameter naming inconsistencies (taskFilePath vs workflowType) and missing chaining IDs in responses will force extra lookup calls.
Copy agent templates from catalog to project.
Copy workflows from catalog to project.
Read current workflow position (read-only). Returns step info and outgoing edges.
Generate a mermaid flowchart diagram for a workflow.
Initialize workflow on task. Idempotent: returns current state if already initialized. Task stays pending until Start() is called.
List catalog contents (workflows and agents).
List available workflows. Filters by source when project workflows exist.
Tool names lack action verbs. 'Init', 'Start', 'Current', 'Next' are weak; should be 'initialize_workflow', 'begin_workflow', 'get_current_step', 'advance_workflow'. LLMs infer intent from names, weak names cause wrong tool selection.
No output schemas documented. LLMs cannot predict what fields are returned (e.g., does Current return step_id, step_name, edges, metadata?). Forces agents to guess and causes context loss between calls.
File path parameters (taskFilePath, filePath) accept arbitrary strings with no validation guidance. No mention of path traversal protection, sanitization, or allowed directories. Security risk for agents writing to unexpected locations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Load workflows at runtime from project or external plugin.
Advance workflow based on step outcome. Returns new step info and outgoing edges.
Get workflow selection dialog for user interaction.
Begin work - advance from start node to first real step. Sets task to in_progress. Requires Init() first.
Parameter descriptions lack constraint details. 'workflowType' has no enum or pattern; 'source' enum is present but others lack validation hints. 'stepId' has no format guidance. Agents will pass invalid values without clear error recovery paths.
No error handling guidance in descriptions. What happens if taskFilePath doesn't exist? If workflowType is invalid? If step has no outgoing edges? Descriptions must tell LLMs what to do next on failure.