Loading ToolBench
One memory for every AI coding agent — plain markdown in your git. Decisions, projects, ledger & board together; recall@5 0.925; 12 engines via install.
samemind provides 10 tools with consistent naming (memory_* prefix) and reasonable descriptions (avg ~80 chars). All tools have input schemas with type definitions and parameter descriptions. However, schemas lack output documentation, error handling guidance is absent, and several parameter descriptions are minimal. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Naming is verb-noun compliant but generic ('memory_' prefix dominates). Parameter constraints (enums, ranges) are missing. No pagination support despite list tools. Descriptions are functional but lack LLM-optimized context (when to use, prerequisites, recovery paths).
Create a new concept/document in the memory bundle
Delete a concept/document from the memory bundle
Retrieve the project board state with status and topics
Retrieve a specific concept/document by ID
Retrieve the event ledger with topics and open failures
Get the current projection status and liveness
List all concepts/documents in the memory bundle with optional filtering
Search and retrieve relevant memories using keyword or semantic search
No output schemas documented. Tools return data but LLMs cannot plan downstream calls or extract fields without trial-and-error. memory_recall, memory_list_concepts, memory_get_concept, memory_get_board, memory_get_ledger, memory_get_status all lack return type documentation.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. memory_delete_concept is DESTRUCTIVE but unmarked. memory_write_inbox and memory_update_concept are WRITE but unmarked. Agents cannot distinguish safe from unsafe operations.
Parameter descriptions are minimal or missing context. memory_list_concepts 'type' and 'tag' lack examples or valid values. memory_recall 'top_k' lacks range (1-100?). memory_update_concept 'frontmatter' is vague ('Frontmatter fields to update'). No constraints prevent invalid input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Update an existing concept/document
Append unstructured content to a single inbox file for later processing
No error handling guidance. Tools provide no recovery hints. If memory_get_concept fails with 'not found', LLM has no suggestion to call memory_list_concepts or memory_recall. If memory_create_concept fails due to duplicate title, no guidance on resolution.
List tools lack pagination. memory_list_concepts accepts 'limit' but no offset/cursor. If bundle contains 1000 concepts, returning all 1000 blows context window. No next_cursor or total_count in response schema.