A desktop AI pet application with tool registry, MCP server integration, and autonomous agent capabilities
MOYU is a C-based agent runtime with 22 tools spanning system monitoring, file/memory management, git operations, and weather. Critical gaps: (1) Most tool descriptions are extremely brief (10-30 chars), well below the 50-200 char LLM-optimized baseline. (2) Input schemas are present but minimal, many tools have empty objects {} with no parameter descriptions. (3) No output schemas documented anywhere. (4) Tool naming is inconsistent: some follow verb_noun (system.now, memory.remember), others use domain.action (git.recent_commits), and several are vague (runtime.explain, file.preview). (5) No error handling guidance visible in any tool. (6) Destructive tools (collection.remove, draft.remove) lack confirmation patterns. (7) Parameters like 'root' and 'path' lack format/constraint descriptions. The codebase shows a lightweight dispatcher (tool.h/tool.c) with basic registration, but tool definitions appear to be hardcoded in builtin.c with minimal metadata.
Add a note to the collection with title and body
List all markdown files in the collections directory
Remove a markdown file from the collections directory
Create a private markdown draft with title and body
List all markdown files in the drafts directory
Remove a markdown file from the drafts directory
Get preview information about a file or directory path
Most tool descriptions are under 30 characters, far below the 50-200 char LLM-optimized baseline. Examples: 'Get current system time' (24 chars), 'Get system idle time' (20 chars), 'Get the currently active foreground application process path' (60 chars but vague). LLMs cannot determine when to select these tools without richer context.
Input schemas are missing or empty for 11 tools (system.now, system.idle, system.foreground_app, system.snapshot, collection.list, draft.list, runtime.explain, runtime.cancel). These tools show Input: {} with no parameter descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
Read-only observation of filesystem contents within authorized root
Recent local commits.
Local branches ordered by age.
Read-only TODO candidates.
Remove entries from memory matching a query
Store text in memory under a specified section
Create a private Markdown draft.
Publish a draft into the configured notes root.
Cancel the current active agent intention
Explain the current runtime agent status, goal, and policy
Get the currently active foreground application process path
Get system idle time in seconds
Get current system time in local and Unix milliseconds format
Collect and format a system snapshot including CPU, memory, disk, and process information
Current weather and a short forecast from Open-Meteo.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing what each tool returns. E.g., does system.snapshot return {cpu: number, memory: number} or {cpu_percent: string, memory_mb: number}?
Destructive tools (collection.remove, draft.remove) lack confirmation or dry-run patterns. No error handling guidance visible. An agent could delete files without safeguards. Per pattern:confirmation-request, irreversible operations should support confirmation.
Parameter descriptions are missing or trivial for 'root' and 'path' parameters across git and filesystem tools. No format constraints, no examples of valid paths, no guidance on absolute vs relative paths. LLMs will guess and pass invalid values.