Loading ToolBench
A workspace-based AI agent platform with playbook execution, task management, and document processing capabilities
Tessera has 23 tools with basic schemas and descriptions, but significant quality gaps prevent higher scoring. Most tools have descriptions (good), but many are generic or lack actionable detail. Parameter descriptions are sparse or missing entirely. No output schemas are documented. Error handling is minimal, tools lack recovery guidance. Tool names follow verb_noun convention reasonably well, but composition issues exist (e.g., 'shell' and 'browser' are catch-all tools combining multiple concerns). Schemas are present but incomplete: many parameters lack type constraints, enums, or validation hints. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are visible. The 'shell' tool (22) is particularly problematic, it accepts arbitrary commands with minimal parameter structure, inviting injection and misuse.
Execute browser automation actions (open, snap, see, click, type, select, back, reload, eval, close).
Ask the user one blocking clarification question in the task UI.
Get available PDF processing capabilities.
Create a new PDF document from blocks (text, images, tables, headings).
Extract text from a PDF document.
Inspect a PDF document to get metadata and page information.
Write a PDF packet manifest for tracking operations.
shell tool is a catch-all combining web-search, web-fetch, Google Workspace, and HubSpot operations. This violates single-responsibility principle and invites injection attacks. Parameter 'command' accepts arbitrary strings with minimal validation.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. Agents must infer structure from trial and error.
Many parameters lack type constraints, enums, or validation hints. E.g., 'operation' in pdf_transform accepts 'split, merge, reorder, rotate' but no enum constraint. 'pages' object parameters lack structure definition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Render PDF pages to images.
Transform PDF documents (split, merge, reorder, rotate).
Validate a PDF document.
Create a Tessera-importable playbook package folder with starter files inside the selected workspace.
Run package-local build/tests and Tessera playbook validation for a workspace playbook package.
Diagnose playbook run issues and provide debugging information.
Execute shell commands for web search, web fetch, Google Workspace operations, and HubSpot integration.
Create and update the task checklist shown in the task UI.
Edit a file inside the selected workspace by replacing text.
Extract readable content from Word, Excel, and PowerPoint files inside the selected workspace.
List files and folders inside the selected workspace.
Run the workspace CLI health check.
Read a text file inside the selected workspace, or extract readable content from supported documents.
Search for text patterns in workspace files.
Write content to a file inside the selected workspace.
Dry-run write probe used by workflow approval steps.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. LLMs cannot distinguish safe read-only tools from destructive ones without explicit hints.
browser tool combines 10 distinct actions (open, snap, see, click, type, select, back, reload, eval, close) in one tool. Should be split into separate tools for clarity and composability.