Recently added

Stop AI coding agents from accidentally emptying your database. Ships an always-on hook that blocks destructive DB commands (DROP, TRUNCATE, DELETE-without-WHERE, migrate:fresh, db:drop, prisma migrate reset, FLUSHALL, docker compose down -v and more) across 15+ frameworks, plus a stack-aware installer for database-level privilege separation and framework boot guards.

Personalized, behavior-aware analysis of where your Claude Code (Max/Pro) usage-limit tokens go, with ranked, low-effort ways to use fewer — computed entirely from your local session logs. Python measures; Claude concludes.

Opt out of telemetry across every locally installed AI CLI and AI-enabled IDE.

A base for vibe-coded projects: required checks on GitHub, a curated set of agent skills, and a generator for new repositories.

Run and steer Tractor pipelines from Codex through MCP.

A valid test plugin for fixture tests. Includes 1 skill and 2 MCP servers.

test-misplaced is an open-source Claude Code plugin indexed from GitHub. Review its manifest and source before installing.

Tools for auditing and maintaining The Workshop's skills, plugins, and distribution boundaries

The complete Workshop toolkit — every skill, agent, methodology doc, and safety hook in one package, including the vault lifecycle, graph, capture, search, sync, and writing workflows. Skills and agents install on Claude Code, Codex, and Cortex Code; the safety hooks execute on Claude Code and Cortex Code. Plan, build, and ship with the full first-party dev workflow.

Socratic thinking partner — sharp questions and decision-sharpening over answers.

Terse senior-staff-engineer voice — answer-first, minimal, expert assumptions. The least verbose persona.

Senior-staff-engineer voice at full depth — reasoning, tradeoffs, and edge cases spelled out.

Momentum-first voice — blunt, bias-to-action, picks a sensible default and moves.

Collaborative pair-programmer voice — brief think-aloud, checks in at decision points.

Product-strategy sounding board and coach persona for a design+PM hybrid at an early-stage startup — decision stress-testing with a steelman duty, influence-case building, prioritization on thin evidence, and verdict-first design critique. Built by persona-builder.

Product-design/UI-UX advisor persona — artifact-first design reviews with severity-tagged findings, named principles, and a stance contract that holds positions against pushback. Built by persona-builder.

The knowledge layer of a public agent-harness reference: 15 dense, opinionated engineering skills covering AWS infrastructure, React/Vite frontends, BFF-on-Lambda backends, CI/CD, the development loop and its quality gates. Written project-agnostically with <project> / <apex-domain> placeholders. This Power ships skills only — the harness's personas, permission hooks and merge gates are deliberately not exported.

A Claude Code harness: 8 subagent personas — two leads that disagree by design (product-and-market, tech) and consolidate one demand, a tool-less scrum-master that ranks the eligible pool and names who acts next in a selection record the main session executes, one fullstack developer, a content pair that argues against one shared ruler at most twice (a content-writer that drafts published prose in the owner's voice and a content-reviewer that may block only on a clause it can quote, both contained the same way as product-lead — neither posts to a public surface directly), one gatekeeper (a quality-assurance that holds the merge gate and reviews both technical delivery and whether the change can break production), and an agents-lead that stress-tests proposed changes to the harness itself and, per ADR-0002, may implement the harness changes it reviews (never merging, never gating an MR) — PreToolUse hooks that mechanically deny the irreversible floor (force-push, terraform apply, rm -rf, secret writes), and 15 generic skills for AWS infrastructure, React/Vite frontends and CI/CD (<project>/<apex-domain> placeholders).

Skills for AI-native development. /assess scores a codebase's readiness for AI agent contributors (0-8 layered contract model) and generates a Codecov-style complexity hotspot SVG plus a colour-blind-safe doc-navigability graph. /huddle runs structured multi-perspective deliberation using Six Thinking Hats with Fibonacci team sizing. /deslop detects and removes the telltale signs of AI writing from prose. /ghsync bulk-clones and keeps in sync every GitHub repo you can access across an org or personal account, for onboarding into a new enterprise. /ghreport is its read-only companion: it reuses ghsync's repo discovery to query each repo's remote state (open PRs, CI on the default branch, security alerts, branch protection) and rolls it into a terminal summary plus a timestamped markdown report of what state an org's repos are in. /skill-forge hardens a skill through judge-panel refinement rounds until it clears a 3-tier promotion gate - a prove-and-promote quality gate that ran on itself. /semantic-compress optimizes an LLM-directed document while preserving what it does, with two transforms gated on the ab-equivalence A/B harness: compress (a local core->pointer pass and an A/B-validated distill loop that produces the smallest behaviourally-equivalent version of a whole document or skill) and directive-clarity (rewrites latent-action instructions - bare negations, facts-not-actions, vague pointers - into directives that name the action, validated by a measured directness gain at zero regression). Also bundles personal workflow commands: /tm (Task Master orchestration), /issues (GitHub-issue marathon - triage open issues then run agent-ready ones to merge with Agent Teams), /fix-pr and /fix-develop (autonomous PR/branch fix loops). /tm, /issues, /fix-pr, and /fix-develop share the marathon and pr-review-merge skills for team orchestration and PR review/merge.

Skills for developing Roslyn analyzers, code fixes, source generators, and diagnostic suppressors.

Provenance-aware memory for Claude Code: a generated MEMORY.md index, per-prompt retrieval of un-loaded memories, and the memsom MCP tools (retrieve, ask, blame, revoke...). Requires `pip install memsom` so the `memsom` / `memsom-mcp` executables are on PATH.

Public Suede skills and a read-only MCP for decisive agent workflows, code review, design, copy, SEO, app shipping, and creator work.

Compound V for Superpowers: triple parallel pre-flight (code archaeology + domain-expert advisor + library/doc validator — Context7 MCP when the session has it attached, WebSearch/WebFetch otherwise), disjoint file partitioning, manifest-driven multi-backend dispatch (Claude + headless Codex + Antigravity + Cursor workers), git-diff scope enforcement, crash-resumable runs, adaptive tier-based routing, epic mode for multi-feature builds, V-memory local-first semantic+lexical recall over docs/superpowers (opt-in pure-python embeddings + a deterministic recall→action bridge), and batched parallel dispatch (Opus default, Sonnet for narrow junior-task carve-out), plus /v:onboard — a project-onboarding command that builds a citation-verified knowledge base + AGENTS.md/CLAUDE.md bridge behind a human gate. Auto-intercepts three transitions: brainstorming → writing-plans → execution; a gated pre-brainstorm recon pass and batched elicitation are guidance the brainstorm follows, not enforced mechanisms.

Verifies signatures on the instruction files Claude Code loads, meaning CLAUDE.md, skills, and agent definitions, before they reach model context. It re-checks a skill bundle before the Skill tool runs it. Sigstore keyless, offline verification against a pinned trust root, no telemetry.

A three-layer development pipeline — project manager, task supervisors, specialist workers — that takes a project from a client's idea to a working product under a written charter, with traceable requirements, independent verification of every report, and escalations classified by reversibility.

Root manifest reached only via fallback. Includes 1 skill.

Claude dialect manifest. Includes 2 skills.

Codex dialect carrier. Includes 1 skill.

A policy-enforcing Streamable HTTP MCP bridge from ChatGPT to local Codex.

An AI-aware design system, read from the code you already have: your coding agent writes each page the way your other pages are built

横屏真人电影感 Seedance 提示词插件，包含精确焦段、表演、分镜、动作特效、室外环境反馈、本地场次记忆和《风起玲珑骨》项目资料。

A non-destructive identity and memory spine for Codex agents.

Agentic workflow orchestration with explicit plans, evidence, and completion gates.

Review agent-authored code you're accountable for but didn't write. /scr:review opens a side-by-side viewer that clusters the diff into semantic groups and annotates what each block changed, so you can judge whether it matches intent and how it was built; your inline comments round-trip back into the session for the agent to iterate on. The LLM directs attention — it never gates.

Skill plus policy directory: role definitions and gate scripts that turn one AI session into a coordinator dispatching specialised workers through Orca

Personal dev pipeline: feat (plan) -> autopilot (build) -> worktree-dev (isolation), plus resume-tailor, multi-agent review, and dsh-based DeepSeek delegation

Drive a Forge issue tracker from the terminal — browse, read, file and comment over the MCP endpoint with no MCP client connected, writing every issue in natural Vietnamese through the bundled vi-natural CLI. Also the global half of the two-level rule arrangement: it guards the shell commands that cannot be undone, stops once before a memory or skill write, and routes every code file a call wrote — through Edit or through the shell — to the project's own eslint, whose rules stay the project's.

Lint edited JavaScript and TypeScript files for maintainable comments and god-file limits with the project's local ESLint installation.

Agent Skill for auditing and improving mobile app UI/UX flows across Flutter, React Native, iOS, and Android.

Claude Code plugin for the mirza-bots fleet: contains the whole engine -- it polls this session's bot on Telegram, stores the fleet's shared history, and exposes reply/read_history/search_history as MCP tools. No daemon.

Anti-sycophantic multi-pass code review: architecture fitness, maintainability, and hidden assumptions

Python type checker and language server (ty by Astral) for code intelligence

Terraform language server for code intelligence and refactoring

Nix language server (nixd) for code intelligence and diagnostics

Scala language server (Metals) for code intelligence, refactoring, and analysis

LLM-friendly packet dissector with MCP server for pcap/pcapng analysis

Persistent memory for Claude Code. Facts are typed, sourced, and retired when they change, so a query returns the current answer rather than every answer it has ever held. Local SQLite in your repository — no API key, nothing uploaded.

OpenTelemetry instrumentation and observability-as-code for any stack

Cut agent-session cost on Claude Code and Codex: cache-expiry prevention, zero-LLM context restore across both tools, cross-tool session handoff, and a real cost dashboard.

Annotated example Agent Plugin: one skill plus one stdio MCP server, packaged in the portable agent-plugins.org format that fleet, Cursor, VS Code, Copilot, Codex and Kiro all load. On the Kubernetes path the server runs in the control-plane pod and the skill reaches sandbox pods through fleet's staged skills tree.