the-claude-factory

Audits Claude skills against the official best practices and gates them: five validation scripts, four independent reviewers, a cold executor and an adversarial verifier give one report with file-and-line evidence and one verdict, and a gate has Claude audit every skill it creates or changes before it finishes.

1 skill · 6 agents

GitHub automation for Claude Code, with GitHub's MCP server or gh: takes pull requests from branch to merge (CI checks, code review comments, fixes), writes PR and issue templates, discussion forms, labels and Dependabot setup, repairs failing GitHub Actions CI, and runs repository audits and cleanup of branches, worktrees and stashes behind a verified backup.

4 skills · 2 commands

Designs, builds and refactors Tailwind CSS v4 (tailwindcss) frontend interfaces with a product designer's craft: UI design, design-system tokens, responsive design and dark mode, every utility class and variant grounded in its exact official documentation page and verified in your project.

2 skills · 4 agents

Writes and reviews Svelte 5 and SvelteKit 3 frontend code by current practice (runes, snippets, Astro and Tailwind CSS projects) and runs the migration from Svelte 4 and SvelteKit 2, checking every change against the official Svelte docs MCP server and autofixer, the Svelte language server (LSP) and your project's own svelte-check.

3 skills · 2 agents

Designs, writes, audits and debugs Claude Code hooks: hook events, matchers and configuration, exit codes and JSON decision control, prompt, agent and async hooks, MCP tool hooks, security and workspace trust, and how mods interact with hooks, checked against the current docs and changelog. Automation you can trust.

1 skill · 1 agent

Writes, edits, reviews, audits, debugs and refactors Claude Code instructions and memory files (CLAUDE.md, CLAUDE.local.md and .claude/rules at user and project scope) against the current loading rules, with a read-only checker and an independent reviewer agent. Documentation that Claude actually follows.

1 skill · 1 agent

Builds and maintains a public Claude Code plugin marketplace on GitHub from the official docs, distilled for plugins that third parties install: marketplaces, manifests, skills, agents, hooks, mods, MCP servers, settings, validation and releases, checked against the latest release notes. Includes the complete plugin CLI reference: every claude plugin and claude plugin marketplace subcommand, /plugin form and session flag, with exact flags, messages, JSON fields, exit codes and minimum versions.

2 skills

Testing for Claude Code plugins before release: installs a copy the way your users receive it in a throwaway config, checks it loads, runs your plugin eval suite against that copy, and reviews your evals so they measure what the plugin adds. Quality you can prove.

2 skills

Keeps a project's memory and documentation in its .claude folder: architecture decisions recorded as ADRs, technical debt tracked in a ledger, and a session handoff that carries each session's state to the next one for continuity.

2 skills

Bash and shell script code intelligence for Claude Code: connects bash-language-server (LSP), formats and lints every script Claude edits with shfmt and ShellCheck using your own configuration, reports the diagnostics, and teaches Claude to trace sourced files and same-named functions before a rename.

1 skill

Designs, builds, reviews, tests and debugs Claude Code mods: panes, bands, function hooks and the hooks module, checking every interface against the current mods API docs and the type declarations your own Claude Code build writes. Plugin development for the Claude Code interface.

1 skill

Makes Claude an expert in Claude Code plugin evals: designs eval suites and benchmarks that prove what a plugin or its skills add, writes cases, graders and mocks, runs and gates them in CI, and reads the results, always checked against the CLI and docs you have today. Testing and quality for plugins.

1 skill

Prompt engineering for Claude: turns a goal, idea or draft into one copy-ready prompt or system prompt tuned to the target Claude model and effort, clarifies intent, checks current official facts, helps with model migration and recommends the setup to run it with.

1 skill

Python code intelligence for Claude Code: connects the basedpyright language server (LSP), lints, formats and type-checks every Python file Claude edits with ruff and basedpyright using your own configuration, and teaches Claude to navigate by symbol, resolve aliases and imports, and fix diagnostics without suppressing them.

1 skill

TypeScript and JavaScript code intelligence for Claude Code: connects your TypeScript 6 language server (LSP), type-checks every file Claude edits with your own tsc under a strictest policy, and teaches Claude to find the program that owns a file and fix compiler diagnostics without unsafe assertions or suppressions.

1 skill

TypeScript and JavaScript code intelligence for Claude Code: connects the native TypeScript 7 language server (LSP), type-checks every file Claude edits with your own native TypeScript 7 compiler (tsc) under a strictest policy, and teaches Claude to find the program that owns a file and fix compiler diagnostics without unsafe assertions or suppressions.

1 skill

Security guardrail for git: blocks Claude's commits and pushes that skip pre-commit hooks or signing (--no-verify, --no-gpg-sign, commit.gpgsign=false), so your hooks, linters and commit signing always run.

A live file explorer panel beside the chat: your project's file tree with git status, diffs and line counts, what Claude reads, writes and commits, search, file previews and your editor one key away. A Claude Code mod for productivity.

Development guard for macOS: stops recursive grep from silently skipping gitignored and binary files, blocks BSD awk's ignored IGNORECASE, and adds the macOS form when a BSD or zsh command fails.

Usage observability for Claude Code: shows, as Claude works, each skill, agent, MCP server, plugin, hook and web tool it uses in the session, how many times and at what time, with an export to Markdown or HTML. A Claude Code mod.

Audit and observability panel for every tool call of the session: what ran, on what, whether Claude or a subagent made it, and whether it finished, failed or was denied by permissions or hooks. A Claude Code mod.

Shows how long each turn took (duration), how many tools it used in the main thread and in subagents, and the tokens it spent, in a status band above the prompt, with usage at a glance and a notification when a long turn ends. A Claude Code mod.