elenchus_start_session
Initiate a new code verification session by collecting initial context, building a dependency graph, and setting up the mediator for adversarial analysis.
Instructions
Start a new Elenchus verification session. Collects initial context, builds dependency graph, and initializes mediator.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| target | Yes | Target path to verify (file or directory) | |
| requirements | Yes | User verification requirements | |
| workingDir | Yes | Working directory for relative paths | |
| maxRounds | No | Maximum rounds before forced stop | |
| verificationMode | No | Verification mode configuration for controlling convergence behavior. Use "fast-track" or "single-pass" for one-shot verification. | |
| differentialConfig | No | Differential analysis configuration. When enabled, only verifies code that has changed since the last verification baseline. | |
| cacheConfig | No | Response caching configuration. When enabled, caches verification results to skip re-verification of unchanged files. | |
| chunkingConfig | No | Selective context configuration. When enabled, chunks files into function-level pieces for more efficient verification. | |
| pipelineConfig | No | Tiered pipeline configuration. When enabled, uses screen→focused→exhaustive verification tiers with auto-escalation. | |
| safeguardsConfig | No | Quality safeguards configuration. Ensures verification quality when using optimizations (caching, chunking, tiered). | |
| dynamicRoleConfig | No | Dynamic role generation configuration. When enabled, generates customized Verifier/Critic roles based on requirements using LLM. | |
| llmEvalConfig | No | LLM-based evaluation configuration. When enabled, uses LLM reasoning for convergence, severity, edge case, and false positive evaluation. |