Loading ToolBench
An agentic legal architecture providing MCP tools for AI agents to collaborate on legal document analysis, workflow management, memory systems, and quality verification. Includes debate board, scoring engine, approval gates, verification loops, and knowledge base integration.
Lavern exposes 64 tools but lacks critical definition infrastructure. Tool names are visible in an allowlist (tool-allowlist.ts) and a label map (toolLabels.ts), but NO input schemas, parameter descriptions, or output schemas are present in the provided source. Tools are referenced by name only, actual MCP tool registration with callTool handlers is not shown. Per the hard rules: missing input schemas → schema score 0; missing descriptions in source → description score 0. The few descriptions that exist (e.g., 'Get the current step in the workflow state machine') are generic and under 50 chars. No evidence of parameter validation, error recovery guidance, or idempotency markers. This is a skeleton implementation.
Advance to next workflow step
Analyze document heading structure
Apply revision guidance to documents
Assess risk factors in legal work
Calculate readability score
Challenge a finding posted on the debate board
Check WCAG accessibility compliance
No input schemas visible for any of 64 tools.
Tool definitions appear only as string names in allowlists and label maps. No explicit MCP tool registration with callTool handlers visible.
Most tools (52 of 64) have no descriptions in source. Only 12 tools in tool-allowlist.ts have brief descriptions (20-40 chars).
No parameter descriptions visible. Tools like 'advance_step' and 'submit_handoff' have no documented parameters, types, or constraints. LLMs cannot infer parameter meaning from names alone.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Compare clauses between documents
Compute readability score
Convert passive voice to active voice
Count passive voice usage in text
Cross-verify work between agents
Check consistency of defined terms
Mark document assembly as complete
Retry document assembly on failure
Start document assembly process
Generate change log from revisions
Get baseline quality scores for document types
Check current progress in workflow
Retrieve the current state of the debate board
Retrieve memory entries
Advance to the next step in the workflow state machine
Get the current step in the workflow state machine
Retrieve handoff summaries from previous workflow phases
Retrieve a specific knowledge base entry by ID
Get the complete workflow history and state transitions
List available knowledge base collections
Load matter-specific memory and context for the current engagement
Query anti-patterns and common mistakes to avoid in legal work
Query institutional memory for firm-wide knowledge and precedents
Query precedent memory for similar past cases and solutions
Search the knowledge base for reference documents and institutional knowledge
Submit a structured handoff summary for phase transitions
Measure sentence length for readability
Analyze visual hierarchy in documents
Merge multiple revision layers
Parse document structure and content
Submit quality scores after document processing
Post a finding to the debate board for agent collaboration
Make a gate decision at approval checkpoints
Calculate pricing for legal engagement
Price risk assessment for engagement
Run quality checks on document or work product
Read and parse a document
Read memory entries
Rebuild PDF bookmarks from document structure
Recalculate readability metrics
Generate session quality metrics report
Resolve a debate on the debate board
Restructure document heading hierarchy
Check color contrast for accessibility
Save a precedent to institutional memory
Score document across multiple dimensions
Score a document across multiple quality dimensions
Search U.S. case law via CourtListener
Search knowledge base for reference documents
Search precedent memory for similar cases
Search document sections
Self-check output for quality and correctness
Compare semantic differences between documents
Simplify sentence structure for readability
Split compound sentences into simpler ones
Store an insight in institutional memory
Run verification on the current workflow step
No error handling guidance. No recovery hints, retryability classification, or actionable error messages documented. Agents will not know how to respond to failures.