AI-first terminal for developers who code, not click. Includes AI chat capabilities with tool use for file operations, terminal commands, web browsing, and widget management.
Wove exposes 29 tools with basic naming and minimal descriptions. Most tool names follow verb_noun convention (capture_screenshot, read_text_file, web_navigate), which is positive. However, descriptions are extremely brief (10-50 chars), providing minimal context for LLM selection. Input schemas are present but sparse: most parameters lack type constraints, enums, or validation guidance. No output schemas are documented. Error handling is absent, no recovery guidance, retryability classification, or actionable error messages visible in the code. Security concerns: file operations (read_text_file, write_text_file, delete_text_file, edit_text_file) accept arbitrary paths with no sanitization hints, risking path traversal. Terminal operations (term_run_command) accept arbitrary commands with no validation. Web operations (web_exec_js) allow arbitrary JavaScript execution. No permission gates or audit trail patterns visible.
Capture a screenshot of a specific widget in the current tab
Close a widget in the current tab
Delete a text file
Edit a text file by replacing specific sections
Generate an image using MiniMax image-01 model
Search for text patterns in files
Invoke a custom skill
Get the owner profile information
Descriptions critically short (10-35 chars). Rubric baseline: 194 chars avg, 34-392 range. Most tools lack context for LLM selection. E.g., 'Read the contents of a text file' (35 chars) does not explain when to use vs grep, or what happens on missing files.
No output schemas documented. Rubric requires: 'Document the output schema. LLMs need to know what fields to expect.' Agents cannot plan downstream calls or extract data without knowing response structure.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | <=2025-11-25 | v2 |
Read the contents of a directory
Read the contents of a text file
Get a structural map of the repository for code exploration
Get the scrollback history from a terminal widget
Run a command in a terminal widget
Send input to a terminal widget
Consolidated utility tool for plans, session history, and project instructions
Capture a screenshot of a web widget
Click an element in a web widget
Execute JavaScript in a web widget
Get console messages from a web widget
Inspect Vue component state in a web widget
Perform a mouse click at coordinates in a web widget
Navigate to a URL in a web widget
Open a new web widget with a URL
Press a key in a web widget
Read HTML content from a web widget
Read text content from a web widget
Perform an SEO audit on a web widget
Type text into an input field in a web widget
Write content to a text file
File operations (read_text_file, write_text_file, edit_text_file, delete_text_file) accept arbitrary paths with no validation hints. No sanitization guidance against path traversal. Rubric: 'Treat all agent-provided input as untrusted. Sanitize against SQL injection, command injection, and path traversal.'
No error handling guidance. Rubric: 'Error responses must tell the LLM what to do next.' No visible recovery patterns, retryability classification, or actionable error messages. Agents cannot self-correct on failures.
Dangerous operations lack confirmation/dry-run patterns. web_exec_js, term_run_command, delete_text_file, write_text_file are irreversible. Rubric: 'Irreversible operations should support a dry-run or confirmation step. Agents make mistakes, a confirm_before_execute pattern prevents catastrophic errors.'