Loading ToolBench
Zero-dependency, fully-reversible context compression for AI agents. Exposes compression, retrieval, and statistics tools via MCP (Model Context Protocol) over stdio.
Three tools with complete schemas and reasonable descriptions. Naming follows verb_noun convention (compress, retrieve, stats). Descriptions are adequate (100-150 chars) but lack actionable context for LLM selection. Parameters are typed but lack constraints (enums, ranges). Output schema is undocumented, LLMs cannot predict response structure. Error handling exists but is generic. No tool annotations (readOnlyHint, etc.). Composition is sound: each tool has one responsibility, and compress/retrieve form a coherent pair.
Compress large tool output, logs, JSON, or file content before reasoning over it. Returns a compact version that preserves errors, outliers, and query-relevant content. If the result contains a [slimctx-ref ...] marker, the full original can be recovered with slimctx_retrieve.
Recover the byte-exact original content for a [slimctx-ref <ref>] marker seen in previously compressed output.
Token savings accumulated in this session.
Output schema not documented. LLMs cannot predict response structure (e.g., does compress() return plain text or JSON with metadata?). All three tools return text content but downstream tool chaining is unclear.
Parameter 'query' in slimctx_compress lacks constraints. Description says 'steers relevance' but does not specify format, length, or examples. LLMs may pass empty strings, very long queries, or malformed input without guidance.
slimctx_retrieve description does not explain what happens if ref is invalid or expired. Error message is generic ('no content for ref'). LLM has no guidance on recovery (retry? ask user? call compress again?).
No tool annotations (readOnlyHint, idempotentHint). All three tools are read-only and idempotent, but this is not declared in the schema. Clients cannot optimize caching or parallelization.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
slimctx_stats description is vague ('Token savings accumulated in this session'). Does not explain when to call it, what the output format is, or how to interpret the numbers.