Loading ToolBench
Git-native persistent memory, multi-chat isolation & cross-session global intelligence for AI coding agents (Antigravity, Cursor, Claude Code). Zero Dolt. Zero Go. Pure TypeScript.
Beadless defines 13 tools with complete input schemas and descriptions. Naming follows verb_noun convention (remember, recall, delete, create, claim, complete, list, save, sync). Descriptions are substantive (100-200 chars typical), explaining WHAT and WHEN. However, output schemas are entirely undocumented, no tool declares what fields it returns, forcing LLMs to guess downstream field names. Parameter descriptions are present but lack format constraints (e.g., no regex for IDs, no min/max for limits). Error handling is not visible in tool definitions. Tool composition is sound (each does one thing), but missing pagination guidance for list tools and no confirmation pattern for destructive operations (delete, sync).
List past recorded chat sessions and conversations across workspaces.
Save or update the current chat session in persistent memory. Stores session title, summary, key decisions, files touched, and tags so it can be recalled in future sessions.
Retrieve the project context briefing, including architecture overview, conventions, critical lessons, and active task queue. Call this at the start of an agent session.
Delete a memory entry by ID (e.g. "mem-1234") or by descriptive search query.
Search across global cross-project memories, all past chat sessions, and Antigravity conversation transcripts. Call this when asked "beadless global eriş" or when looking for solutions, decisions, or context discussed in other chats/sessions.
Scan and index recent Antigravity conversation transcripts into beadless global memory.
No output schemas documented for any tool. LLMs cannot infer what fields are returned, forcing guesswork on downstream field names and types. This violates pattern:tool and pattern:response-shaper.
Destructive tools (beadless_delete, beadless_global_sync) lack confirmation or dry-run patterns. Agents can permanently delete memories or sync transcripts without user approval, risking data loss.
List tools (beadless_task_list, beadless_chat_list) lack pagination parameters (limit, offset, cursor). No guidance on result caps. Large result sets will blow context windows.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 59 | 2026-07-28+ | v2 |
Search past project decisions, lessons, or architectural context using fuzzy search. Call this before making major architectural changes or when debugging unfamiliar code.
Store a project decision, lesson learned, architectural choice, or gotcha in git-native memory. Survives across agent sessions and git branches.
Capture a snapshot of the current workspace state, modified/untracked files, and git status into persistent memory.
Claim a task to signify an agent is actively working on it. Fails if dependencies are not yet completed.
Mark a task as completed with an outcome summary. Automatically unblocks downstream dependent tasks.
Create a new task with optional dependencies (blockedBy). Use this to decompose complex work into trackable steps.
List project tasks. Can filter by status or query only unblocked (ready) tasks.
Parameter descriptions lack format constraints. 'limit' has no min/max (default 5-6 but unbounded). 'queryOrId' in beadless_delete lacks format guidance (ID pattern vs free-text query ambiguity).
No error handling guidance visible in tool definitions. No recovery hints (e.g., 'If task not found, call beadless_task_list first'). Agents cannot self-correct on failures.