Loading ToolBench
L.O.G. — Latent Orchestration Gateway: Privacy-first middleware for AI agent ecosystems. Implements PII dehydration, rehydration, archiving, distillation, and hysteresis management for agent-facing tool interfaces.
LOG-mcp has 8 tools with complete input schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (log_dehydrate, log_rehydrate, log_distill, log_archive_session, log_search_archives, log_archive_gnosis, log_prune_hysteresis, log_vault_status). Descriptions are present but brief (avg ~80 chars, below the 194-char baseline). Parameter descriptions exist but lack detail on constraints, formats, and valid ranges. No output schemas documented. Error handling and recovery guidance absent. Security concerns: tools accept sensitive data (messages, text) without explicit PII handling guarantees in descriptions. Composition is reasonable, each tool has a single responsibility, but chaining is unclear (e.g., does log_archive_session return a session_id for later retrieval?).
Save a permanent lesson learned to the gnosis directory.
Archive a conversation session to the Vault's local filesystem.
Strip PII from text and replace with LOG_IDs. Always call this before sending data to external agents.
Distill a conversation into a semantic summary (Working-Fiction). Strips fluff, extracts key context.
Run the garbage collector. View or modify memory tier status.
Swap LOG_ID placeholders back to real values for human-readable output.
No output schemas documented. Tools return data but LLMs cannot predict field names, types, or structure. Breaks downstream tool chaining and forces agents to guess at response format.
Parameter descriptions lack constraint details. 'messages' array has no format spec (expected {role, content, timestamp}?). 'limit' has no min/max bounds. 'action' enum is documented but 'new_tier' enum values not explained.
No error handling or recovery guidance. Tools do not document failure modes (e.g., what if archive write fails? What if search returns no results?). Agents have no guidance on retry strategy or fallback.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
Search the Vault's archive index by topic, summary, or tags.
Get current Vault status: storage stats, entity count, tier distribution.
Tool name 'log_prune_hysteresis' is unclear. 'Prune' + 'hysteresis' + 'garbage collector' is jargon-heavy. Rename to 'log_manage_memory_tiers' or 'log_tier_promote' for clarity.
Descriptions are too brief (avg 80 chars vs 194-char baseline). Missing WHEN to use each tool and dependencies. E.g., log_rehydrate description does not say 'Call after log_dehydrate' or 'Requires prior dehydration context'.