Loading ToolBench
Self-contained immutable local digital library and intelligence runtime with web interface, search, and document ingestion capabilities
This server has severe definition quality gaps. Of 20 tools, most lack proper input schemas visible in source code. Tool descriptions exist but are often domain-specific jargon-heavy (e.g., 'SPRE × CLCE × PhysLing evaluation') without explaining WHEN to use them or WHAT they return. Parameters lack type definitions in visible schemas. No output schemas documented. Error handling and recovery guidance absent. The codebase mixes Python (webapp.py, review_applicability.py) and JavaScript (workers/, tools/) with inconsistent tool registration patterns, some tools appear inferred rather than explicitly registered with MCP schemas.
Build a COLD-MULTI-SHELF-1.0 export pack with lockset, manifest, registry, and foldlock
Rule-based domain and subject classification for library tree placement. Returns multi-placement paths with main/sub/micro hierarchy
Classify whether a document has claim-structure for CLCE (Claim-Layer Comparative Evaluation)
Comprehensive applicability classification across all evaluation components (SPRE, CLCE, PhysLing, Truth Formula)
Classify whether a document has physics-evaluable claims for PhysLing (Physics-Linguistics evaluation)
Classify whether a document has provenance signals for SPRE (Survival Provenance Record Evaluation)
No visible input schemas with type definitions. Tools list parameters (e.g., 'q', 'file', 'title', 'body') but source code does not show JSON Schema definitions with 'type', 'required', or 'properties' fields.
Descriptions rely on domain jargon without explaining WHEN to use tools or WHAT they return. E.g., 'classify_spre_applicability' says 'Classify whether a document has provenance signals for SPRE (Survival Provenance Record Evaluation)', an LLM unfamiliar with SPRE/CLCE/PhysLing cannot determine when to invoke this. No output schema documented for any tool.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Classify whether a document qualifies for Truth Formula evaluation (coverup mode or ordinary subset)
Collect library statistics from packed index (views, downloads, by_repo, by_branch, by_fork, github stats, record counts)
Hash all core documentation files in the repository
Ingest a file into the library with structure verification, SPRE × CLCE × PhysLing evaluation, Bayesian scoring, and document hash-chain
Research assistant chat interface for asking about filed records. Not sovereign, not operator, cannot change scores
Hot-path read of packed library index from KV (one get, never list). Returns cached or fresh index
Search library records by query string
Compute SHA256 hash of a file
Verify a file's SHA256 hash against a COLD-MULTI-SHELF pack (lockset tip or manifest)
Verify SHA256SUMS file against actual file hashes in directory
Write complete COLD-MULTI-SHELF airgap pack: export files, checklists, attestation, restore drill, SHA256SUMS, verify script, and tar archive
Write COLD-MULTI-SHELF export pack to directory (lockset.json, manifest.json, registry.json, export.json, foldlock.json, ingest-as-receipt.json)
Write SHA256SUMS file for a directory of files
Write verify-airgap.sh shell script for Plane C airgap verification (SHA256SUMS check, no invented DOI, no AZ-GEN live ICANN publish)
No error handling or recovery guidance. Tools like 'ingest_file' (WRITE risk) and 'writeAirgap' (WRITE risk) have no documented error cases, validation rules, or guidance for LLM recovery.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. E.g., 'search_records' returns results but no schema shows whether results include 'id', 'title', 'author', or pagination fields.
Tool definitions appear inferred from file paths and parameter lists rather than explicitly registered with MCP schemas. Source code shows function implementations (e.g., sha256File in tools/cold_shelf/index.mjs) but no visible MCP tool registration, manifest, or schema binding.