Loading ToolBench
A collection of 53 Claude Code plugins including agents, skills, hooks, and commands for various development tasks
This server has severe definition quality gaps. Tool names lack action verbs (check-assumptions, scan-speculative, load-context are descriptive but not verb-noun patterns). Descriptions are present but generic and lack LLM-optimization guidance. Input schemas are inconsistent: some tools have no input schema at all (add-timestamp, load-context), others have partial schemas with missing type information. Parameters lack descriptions in most cases. No output schemas documented. Error handling is minimal, tools exit with codes but provide no recovery guidance. The server appears designed as bash hooks rather than production MCP tools, evidenced by reliance on environment variables and lack of structured error responses.
Adds current timestamp to user prompts to provide Claude with time context
Warns about assumption language and Canvas context detection - verifies course/assignment/grade details via Canvas MCP tools
Creates a task folder with timestamp, device name, and README template in OneDrive projects directory
Loads project context on session start including CLAUDE.md, git branch, uncommitted changes, and project type detection
Scans response for speculative language patterns and warns if unverified claims are detected
Validates bash commands and blocks dangerous operations like rm -rf, credential leaks, and suspicious curl commands
Validates file write operations and blocks writes to sensitive files and system directories
Two tools (add-timestamp, load-context) have empty input schemas ({}).
Tool names do not follow verb_noun convention. Names like 'check-assumptions', 'scan-speculative', 'load-context' are descriptive but lack action verbs (get_, create_, update_, validate_). LLMs rely on verb-first names to infer intent.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract results without knowing return structure. Descriptions mention JSON output (e.g., scan-speculative returns {result, reason}) but schemas are not formally declared.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2026-07-28+ | v2 |
Parameter descriptions are missing or minimal. E.g., validate-bash has 'command' param with description 'Bash command to validate' (18 chars, below 20-char threshold). validate-write has 'file_path' and 'filePath' (duplicate params with inconsistent naming) with minimal descriptions.
Error handling provides exit codes (0=allow, 2=block) but no structured error messages to guide LLM recovery. Stderr messages like 'Blocked: Dangerous command pattern detected' are human-readable but not machine-parseable. No recovery guidance (e.g., 'Try a safer alternative').