Skill UI

Analyzes Claude Code and Codex usage logs to surface cost drivers such as low cache hit rates, unpaid cache writes, top-tier sub-agents, reasoning share, and bloated sessions, with evidence and what-if savings to help cut AI coding spend.

This tool provides a unified cost view for AI agents like Claude Code, Codex, and Grok. It reads local logs to calculate spend, token usage, and cache hit ratios per provider and model. It flags unpriced models and supports filtering by time range, project, and provider. Output can be JSON or Markdown. All data stays local, ensuring privacy. Ideal for auditing AI usage costs and comparing provider pricing.

A disciplined approach for diagnosing hard bugs and performance regressions. This skill guides users to build tight feedback loops, reproduce issues, and minimize scenarios before hypothesizing. It emphasizes redacting secrets, creating deterministic tests, and avoiding premature assumptions. Use when debugging complex failures or performance issues in software development.

This tool generates backlink ideas by analyzing competitor backlink examples or exports. It helps SEO professionals identify link-building opportunities within their niche by examining competitor strategies. Users provide niche details and competitor data, and the tool suggests actionable prospects and next steps for outreach.

Bifrost is an open-source, self-hosted AI gateway that unifies over 20 LLM providers behind a single OpenAI-compatible API. It supports provider failover, weighted load balancing, virtual keys with budgets, semantic caching, and observability. Ideal for applications needing multi-provider access, key management, and traffic governance without modifying existing SDKs.

Commit Craft is a Git workflow tool that organizes mixed changes into logical, reviewable commits before pushing. It blocks secrets like .env files, lints commit messages against standards, and ensures each commit builds independently. Using Python scripts, it plans series, gates secrets, and validates messages without modifying the repository directly, enhancing code quality and security.

Validates cosine similarity claims over embeddings or hidden states by measuring each side's anisotropy, centring by the population mean, and scoring matched pairs against a permuted floor before any number is reported.