Loading ToolBench
A collection of MCP server configurations for Loki, Telegram, and Transmission services, along with various hooks and skills for Claude Code integration
This collection mixes hook scripts and skills with inconsistent quality. Most tools lack proper input schemas, parameters are documented in descriptions but not in formal JSON Schema with types. Descriptions vary widely: some are detailed (worksync-append, session-monitor), others are vague (arrays, pairwise). No output schemas are documented. Tool names are inconsistent: some use verb_noun (worksync-append), others are nouns (session-monitor, arrays). Error handling is minimal, most scripts exit silently on edge cases rather than returning actionable error messages. Security concerns: worksync-append and session-monitor write to user directories without validation. No tool annotations (readOnlyHint, destructiveHint) despite clear WRITE vs READ_ONLY risk levels.
Main-effects analysis of a finished Taguchi run sheet. Computes level means, deltas, signal-to-noise ratios, and permutation p-values. Supports numeric and binary outcomes, covariates, and multiple goals (minimize, maximize, target).
Taguchi orthogonal arrays (L4, L8, L9, L12, L18) with pairwise balance verification. Arrays are checked at import time to ensure every pair of columns has equally distributed level combinations.
PreToolUse hook that refuses Write/Edit/MultiEdit on .md files when prose paragraphs span more than one line. Detects hardwrapped paragraphs while exempting code blocks, lists, tables, blockquotes, and badge/image rows.
PostToolUse hook that runs golangci-lint on Go files and reports linting issues in JSON format. Skips if no golangci-lint config found.
Greedy pairwise covering array generator with restarts and constraints. Generates a markdown run sheet where every reachable pair of factor levels appears in at least one row. Supports --exclude constraints and deterministic output.
macro-claude session monitor hook. Tracks Claude Code session state (idle, working, waiting, error) and writes compact JSON status files for macropad plugin polling. Handles events: SessionStart, SessionEnd, UserPromptSubmit, PreToolUse, PostToolUse, Stop, StopFailure, Notification.
Input schemas lack formal JSON Schema type definitions. Parameters documented in descriptions only (e.g., lint-go's 'file_path' has no 'type' field in schema). This forces LLMs to infer types and violates pattern:constrained-input.
No output schemas documented. Tools return unstructured text or JSON without declaring field types or structure. LLMs cannot plan downstream calls or extract data reliably.
Error handling is silent or minimal. Scripts exit with code 0 on errors (e.g., lint-go, check-md-hardwrap) without returning actionable error messages. Violates pattern:recovery-guide.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Guard for commit message trailers. Detects and reports missing Signed-off-by trailers, incorrect Claude-Session lines, and incorrect Assisted-by values. Blocks publishing commands when defects remain. Handles PostToolUse (report) and PreToolUse (block) events.
Append an entry to today's work-sync note. Supports types: pr, issue, merge, research, ops, checkpoint, note. Deduplicates exact entries and GitHub URLs already logged today.
PostToolUse hook that detects GitHub-facing gh commands and appends them to today's work-sync note. Matches: gh pr create, gh issue create, gh pr merge, gh issue close.
Tool names lack consistent verb_noun convention. 'arrays' and 'pairwise' are nouns; 'session-monitor' is ambiguous (noun or verb?). 'lint-go' uses verb but is unclear vs 'check-go-lint'. Violates pattern:tool naming baseline.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. worksync-append and session-monitor are WRITE operations; lint-go and analyze are READ_ONLY. Missing annotations prevent LLMs from reasoning about side effects.