Loading ToolBench
Plugin marketplace and hook system for AlphaAgent autonomous coding assistant. Provides pre-commit hooks for git operations, package manager validation, project context injection, and anti-pattern detection.
This server has critical definition quality gaps. All 6 tools lack proper input schemas, parameters are documented in descriptions but not in structured JSON Schema format with type definitions. Tool names use hyphens (block-master-commits) instead of verb_noun convention (block_commits). Descriptions are present but generic (50-100 chars), lacking WHEN/WHY context. No output schemas documented. No error handling guidance. The tools are git/npm hooks that read stdin and exit with codes, but the MCP tool interface is not properly defined. Tools appear to be shell scripts invoked via STDIO, not MCP-native tools with structured schemas.
Prevents direct commits to protected branches (main, master, production, develop). Only allows commits on feature branches.
Detects the project's package manager from lock files and blocks npm commands when pnpm, yarn, or bun should be used instead.
Detects project configuration and provides it to Claude. Eliminates redundant package.json reads and npm vs pnpm confusion. Returns context summary including package manager, git branch, repository name, and technology stack.
Blocks git commits that claim completion without verification evidence. Prevents 'Optimism Debt' where agents claim tasks are complete without providing proof (git diff, build status, test results).
Lightweight validation that runs after every subagent completes. Checks for common anti-patterns WITHOUT running tests (fast). Validates: no backup/temp test files committed, no console.log in production code, no staged backup files.
Validates that completion claims match reality: test count claims match actual test results, 'All tests passing' verified, file modification claims verified. Prevents false claims like '127/127 tests passing' when actually 120/127.
No input schemas defined. All 6 tools lack JSON Schema with type definitions. Parameters documented in prose only (e.g., 'command' and 'cwd' strings), not in structured schema format.
Tool names violate verb_noun convention. Names use hyphens (block-master-commits, block-wrong-package-manager) instead of underscores and action verbs (block_commits, validate_package_manager). LLMs parse intent from names, these are ambiguous.
No output schemas documented. Tools return exit codes and stderr messages, but no structured response format is defined for MCP clients. LLMs cannot plan downstream calls without knowing response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 24 | 2026-07-28+ | v2 |
Descriptions lack WHEN/WHY context. Descriptions state WHAT (e.g., 'Prevents direct commits') but not WHEN to call or WHY it matters. Average length ~60 chars; baseline is 194 chars. Missing dependency hints and recovery guidance.
No error handling guidance. Tools exit with code 2 to block operations, but no structured error response tells the LLM what to do next (retry, ask user, abort). Exit codes are shell semantics, not MCP error contracts.