Loading ToolBench
Identity-aware context compression daemon for LLM applications. Compresses verbose identity/context into compact key=value format and detects redundant content reinforcement.
Engram's 5 tools have complete input schemas with proper JSON Schema structure and type definitions. However, descriptions are minimal (10-60 chars, well below the 194-char baseline), parameter descriptions are sparse or missing, and output schemas are entirely undocumented. Tool naming is reasonable (verb_noun pattern: derive_, compress_, check_, get_, generate_) but descriptions lack the WHAT/WHEN/WHY guidance LLMs need for selection. No error handling guidance, no recovery paths, and no documentation of side effects (e.g., does generate_report write files?). The server appears to be a daemon adapter for identity compression, but the tool interface does not expose enough context for agents to use it effectively.
Check content for redundant identity reinforcement against the active codebook.
Compress identity dimensions into compact key=value format.
Derive a codebook from CLAUDE.md content. Extracts structured identity dimensions via pattern matching.
Generate a markdown token savings report.
Get current session token accounting statistics.
Tool descriptions are 10 - 60 characters, far below the 194-char baseline. 'Derive a codebook from CLAUDE.md content.' lacks WHEN to call it, what it returns, and how it differs from check_redundancy. LLMs cannot reliably select the right tool.
Output schemas are completely undocumented. Agents do not know what fields derive_codebook returns, what structure compress_identity produces, or how to chain results into downstream calls. This violates the documented output schema requirement.
Parameter descriptions are missing or trivial. 'content' in derive_codebook says 'The CLAUDE.md text content' but does not explain format, size limits, or expected structure. 'dimensions' in compress_identity lacks any description. LLMs cannot infer parameter meaning from names alone.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
No error handling guidance. If derive_codebook fails to parse CLAUDE.md, or generate_report cannot write to savingsLogPath, agents receive no recovery hints. Error responses must tell LLMs what to do next.
Side effects undocumented. generate_report appears to write files (savingsLogPath parameter), but the description does not state this. Agents need to know which calls are safe to retry and which have irreversible consequences.