# Agent Toolkit

> 🛠️ Composable AI agent toolkit — skills, agents, loops, and MCP templates for Claude Code, Cursor, OpenCode, Copilot, Windsurf, and Pi

- **Type:** MCP server
- **Install:** `agentstack add mcp-ulises-jeremias-agent-toolkit`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ulises-jeremias](https://agentstack.voostack.com/s/ulises-jeremias)
- **Installs:** 0
- **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [ulises-jeremias](https://github.com/ulises-jeremias)
- **Source:** https://github.com/ulises-jeremias/agent-toolkit
- **Website:** https://ulises-jeremias.github.io/agent-toolkit

## Install

```sh
agentstack add mcp-ulises-jeremias-agent-toolkit
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# agent-toolkit

> **New here?** Start with [Getting Started](docs/GETTING_STARTED.md) — install → doctor → first skill.

**Composable AI agent capabilities for every major coding assistant**

[](https://github.com/ulises-jeremias/agent-toolkit/actions/workflows/validate.yml)
[](https://github.com/ulises-jeremias/agent-toolkit/actions/workflows/mega-linter.yml)
[](LICENSE)
[](https://github.com/ulises-jeremias/agent-toolkit/releases/latest)
[](https://discord.gg/bR5VyATgka)
[](https://github.com/vercel-labs/skills)
[](https://agent-plugins.org)

[](https://www.npmjs.com/package/agent-toolkit-cli)
[](https://www.npmjs.com/package/agent-toolkit-cli)
[](https://pypi.org/project/agent-toolkit-cli/)
[](https://pypi.org/project/agent-toolkit-cli/)
[](https://aur.archlinux.org/packages/agent-toolkit-bin)
[](https://github.com/ulises-jeremias/homebrew-tap)
[](https://github.com/ulises-jeremias/agent-toolkit/pkgs/container/agent-toolkit)

[](https://github.com/ulises-jeremias/agent-toolkit/stargazers)
[](https://github.com/ulises-jeremias/agent-toolkit/commits/main)
[](https://github.com/ulises-jeremias/agent-toolkit/graphs/contributors)

[](profiles/claude-code/)
[](profiles/cursor/)
[](profiles/opencode/)
[](profiles/copilot/)
[](profiles/windsurf/)
[](profiles/pi/)
[](https://developer.meta.com/ai/products/muse-code/)

[Documentation](docs/) ·
[Quick Install](#-quick-install) ·
[Skills](#%EF%B8%8F-skills--catalog-via-agent-toolkit-inventory) ·
[Agents](#-agent-personas) ·
[Loops](#-loop-engineering) ·
[Swarm](#-swarm-orchestration) ·
[Contributing](#-contributing)

---

## ✨ What is agent-toolkit?

**agent-toolkit** is a modular collection of skills, agent personas, MCP configuration templates, and loop engineering patterns that work across all major AI coding assistants. Instead of maintaining separate prompt libraries for each tool, you keep **one source of truth** and deploy exactly what each tool needs.

One toolkit. Any coding assistant. Zero duplication.

```bash
# Native V CLI (any channel below), then:
agent-toolkit install
agent-toolkit doctor
```

---

## Key Concepts

  
    
      🛠️ Skills
      Reusable capability units (SKILL.md) that teach an agent how to do a job — delivery workflows, forge CLIs, design, data, ops.
      
      Skills across 14 domains — live count via agent-toolkit inventory · Browse skills/.
    
    
      🤖 Agents
      Personas that constrain how the AI works in a session — review, plan, architect, fix CI — without rewriting your prompts each time.
      
      Personas under agents/, compiled into each target's native format — see agent-toolkit inventory.
    
  
  
    
      🔄 Loops
      Recurring agentic workflows with mutation-safety Stages (L1 observe → L2 controlled → L3 high autonomy).
      
      agent-toolkit loop run daily-triage · templates in loops/ — see agent-toolkit inventory
    
    
      📦 Packs
      Solution bundles that combine skills, agents, and loops for a team context (OSS maintenance, engineering workflow, delivery discipline).
      
      Declared under packs/; load with your workspace tooling.
    
  
  
    
      🧩 Plugins
      Product distributions for Claude Code / Cursor marketplaces — core, agents, forge — compiled from one catalog.
      
      Source of truth: distributions/products.yaml
    
    
      🔗 MCP
      Provider registry + ready templates (GitHub, Slack, Notion, Linear, Figma, ClickUp, Chrome DevTools) emitted into target-native MCP configs.
      
      7 providers · mcp/registry/ · mcp/templates/ · agent-toolkit mcp
    
  

---

## 🚀 Quick Install

**Recommended:** the product CLI is the **native V binary**. PyPI/`uv` is a thin launcher over that binary ([ADR-021](docs/adrs/ADR-021-pypi-binary.md)).

```bash
# GitHub Release — native V binary + SHA256SUMS (v1.11.0+)
# https://github.com/ulises-jeremias/agent-toolkit/releases/latest

# Homebrew
brew tap ulises-jeremias/homebrew-tap && brew install agent-toolkit

# AUR (Arch) — native V binary; not the Python AUR package
yay -S agent-toolkit-bin

# PyPI launcher (execs bundled V; ADR-021)
uv tool install 'agent-toolkit-cli>=1.11.0'
uvx --from 'agent-toolkit-cli>=1.11.0' agent-toolkit install

# npm
npm i -g agent-toolkit-cli

agent-toolkit install    # auto-detects Claude, Cursor, OpenCode, Windsurf, Pi, Copilot
agent-toolkit doctor     # verify everything is set up
```

→ Full walkthrough: [docs/INSTALLATION.md](docs/INSTALLATION.md)

### Advanced install methods

Use these only when the primary CLI flow above does not fit your setup.

Claude Code plugin marketplace — native plugins for Claude Code only

```
/plugin marketplace add ulises-jeremias/agent-toolkit
/plugin install agent-toolkit-core@agent-toolkit
/plugin install agent-toolkit-agents@agent-toolkit
/plugin install agent-toolkit-forge@agent-toolkit
```

Cursor plugins — Cursor IDE and Cursor Agent CLI

Native plugins from [`.cursor-plugin/marketplace.json`](.cursor-plugin/marketplace.json):
`agent-toolkit-core`, `agent-toolkit-agents`, `agent-toolkit-forge`.

**Cursor IDE**

1. Open **Customize** in the sidebar (or Command Palette → *Cursor: Open Plugin Marketplace*).
2. Import the marketplace repo: `https://github.com/ulises-jeremias/agent-toolkit`
3. Install the plugins you need (`agent-toolkit-core` is the baseline).

User-scoped installs sync to Cursor Agent CLI sessions automatically.

**Cursor Agent CLI**

```bash
# Interactive — browse / install from the Marketplace tab
cursor-agent
# then type: /plugin
```

Load a local plugin directory for one session (useful while developing):

```bash
cursor-agent --plugin-dir ./plugins/agent-toolkit-core
cursor-agent --plugin-dir ./plugins/agent-toolkit-agents
cursor-agent --plugin-dir ./plugins/agent-toolkit-forge
```

**Local / offline**

```bash
# Symlink into Cursor's local plugin dir, then reload the window
mkdir -p ~/.cursor/plugins/local
ln -s "$(pwd)/plugins/agent-toolkit-core" ~/.cursor/plugins/local/agent-toolkit-core
ln -s "$(pwd)/plugins/agent-toolkit-agents" ~/.cursor/plugins/local/agent-toolkit-agents
ln -s "$(pwd)/plugins/agent-toolkit-forge" ~/.cursor/plugins/local/agent-toolkit-forge
```

→ [Cursor plugins docs](https://cursor.com/docs/plugins) · [Plugin marketplace guide](docs/wiki/Plugin-Marketplace.md)

npx skills — Agent Skills standard (skills only, no agents/loops)

```bash
npx skills add ulises-jeremias/agent-toolkit -g
```

Homebrew / AUR / GHCR — system packages and container

Formulas live in dedicated repos ([homebrew-tap](https://github.com/ulises-jeremias/homebrew-tap), [aur-packages](https://github.com/ulises-jeremias/aur-packages)); release workflows notify them on tag. The container image ships the same GitHub Release V binary.

```bash
brew tap ulises-jeremias/homebrew-tap && brew install agent-toolkit
yay -S agent-toolkit-bin   # Arch Linux (AUR) — GitHub Release V binary; not the Python AUR package
npm i -g agent-toolkit-cli # optionalDependencies platform packages
docker pull ghcr.io/ulises-jeremias/agent-toolkit
```

Git clone + install script — offline or pinned commit

```bash
git clone https://github.com/ulises-jeremias/agent-toolkit ~/.agent-toolkit
./make.vsh install-cli    # canonical V binary → ~/.local/bin (or --prefix=…)
agent-toolkit install
```

Prefer `./make.vsh install-cli` or a release channel above ([ADR-007](docs/adrs/ADR-007-install-sh-deprecation.md) removed the deprecated `scripts/install.sh` wrapper).

Manual per-tool copy — full control over paths

```bash
git clone https://github.com/ulises-jeremias/agent-toolkit ~/.agent-toolkit

# Claude Code — reference in .claude/settings.json
# Cursor — copy profiles/cursor/rules/*.mdc → .cursor/rules/
# OpenCode — copy profiles/opencode/ → ~/.config/opencode/
# Copilot — copy profiles/copilot/copilot-instructions.md → .github/
# Windsurf — copy profiles/windsurf/ → ~/.codeium/windsurf/
# Pi Agent — copy profiles/pi/skills/ → ~/.pi/agent/skills/
```

Per-tool steps: [docs/INSTALLATION.md#manual-install](docs/INSTALLATION.md#manual-install)

---

## 🖥️ Supported Tools

| Tool | Type | What's deployed |
|------|------|-----------------|
| **Claude Code** | Plugin + CLI | Plugin manifest, skill references, settings |
| **Cursor** | Plugin + IDE + Agent CLI | Marketplace plugins (`.cursor-plugin/`), `.mdc` rules via profile |
| **OpenCode** | TUI | System prompt overlays, agent configs |
| **GitHub Copilot** | IDE | `copilot-instructions.md` with domain selection |
| **Windsurf** | IDE | Rules and memory files via Cascade |
| **Pi Agent** | Agentic harness | Skills and loop templates in Pi's native format |
| **Muse Code** | CLI | Agent Skills under `~/.config/muse/skills/` |

---

## 🛠️ Skills — catalog via `agent-toolkit inventory`

All skills use `SKILL.md` frontmatter only — no `skill.json` required. Fully compliant with the [Agent Skills spec](https://github.com/vercel-labs/skills). Live counts: `agent-toolkit inventory` / `catalogs/skill-catalog.yaml` (source of truth, not README badges). Marketplace plugins ship a subset (core + forge); `agent-toolkit-complete` is the full catalog.

| Domain | Key Skills |
|--------|------------|
| 🧠 `core` | assistant, dev-companion, workspace, project, onboarding |
| 🚀 `delivery` | adr, bug, epic, development-workflow, planning, prd, user-story, work-item |
| 🎨 `design` | figma-implement-design, figma-code-connect-components, frontend-design |
| ⚡ `forge` | github-cli-workflow, gitlab-cli-workflow, gh-fix-ci, worktree |
| 🔗 `integrations` | slack-cli, slack-assistant, linear, clickup-cli, mcp, jira-*, confluence-* |
| 📊 `data` | dbt-validation, snowflake-validation |
| 🔧 `tooling` | jupyter-notebook, playwright-cli, herdr, inventory |
| 🛡️ `ops` | triage, docs-generator, llm-cost-advisor, swarm |
| 🔄 `loops` | loop-runner (see [Loop Engineering](#-loop-engineering) for 10 templates) |
| 🔐 `agentic-security` | threat-modeling, owasp-agentic-review, mcp-audit |
| ☁️ `cloud` | cloud-design-patterns, aws-well-architected-review |
| 🏛️ `architecture` | architecture-diagram, c4-model |
| ♿ `accessibility` | review |
| ✅ `quality` | megalinter, megalinter-setup, megalinter-check, megalinter-fix, codeql, blast-radius |

Browse the full catalog: [`catalogs/skill-catalog.yaml`](catalogs/skill-catalog.yaml) · **membership matrix** [`docs/SKILL_PRODUCT_MATRIX.md`](docs/SKILL_PRODUCT_MATRIX.md) (`scripts/generate-skill-matrix.vsh --check` in CI) · regenerate with `./scripts/validate-skills.vsh` (CI) and inspect live inventory via `agent-toolkit inventory` (counts shown in badges above are generic — use inventory for accurate numbers)

### Loading skills in Claude Code

```jsonc
// .claude/settings.json
{
  "skills": [
    "ulises-jeremias/agent-toolkit/skills/core/assistant",
    "ulises-jeremias/agent-toolkit/skills/delivery/workflow-generic-project",
    "ulises-jeremias/agent-toolkit/skills/delivery/bug"
  ]
}
```

---

## 🤖 Agent Personas

Tool-agnostic agent persona definitions in `agents/` (live count: `agent-toolkit inventory` / `catalogs/agent-catalog.yaml`). Any supported AI coding assistant can import these via its profile config. Canonical taxonomy: [`docs/AGENT_TAXONOMY.md`](docs/AGENT_TAXONOMY.md).

### Holistic roster — the daily set (11)

Every skill's `holistic_owner` in `capabilities/skills/registry.yaml` is one of these. Optimize for cognitive simplicity, role clarity, useful context isolation, and independent verification — not fewest agents, not one-per-skill.

| Persona | Responsibility | Main skill domains |
|---------|---------------|--------------------|
| 🤝 `assistant` | **Orchestrator** — intent → context → proportional delegation → synthesis | `core/*`, discovery, `output-handshake` |
| 📋 `planner` | Decomposition, PRD/TRD, work items, estimation, capacity | `delivery` (11 inc. `planning`, `project-assessment`, `workflow-generic-project`) |
| 🏗️ `architect` | System design, tradeoffs, C4, ADRs/TRDs, cloud patterns | `architecture` (2), `cloud` (2), `delivery` (adr/trd/decision-log/technical-assessment), `tooling/mermaid` |
| 🎨 `designer` | Visual direction, UX, Figma, design system, a11y — contextual routing | `design` (10), `accessibility/review` |
| 🔨 `implementer` | Feature/bug/refactoring delivery, build/test loop, docs generation | `delivery/task`, `ops/docs-generator` |
| 🔍 `reviewer` | Independent quality/craft, change-safety, anti-slop (`blast-radius`, `deep-review`, `deslop`, `unslop`) | `quality` (4) |
| 🧪 `qa-engineer` | Behavioral verification, lint gates, browser automation, E2E, bug triage | `quality/megalinter*` (4), `tooling/playwright-cli`, `tooling/chrome-devtools`, `delivery/bug` |
| 🔒 `security-engineer` | App + agentic hardening, threat modeling, supply-chain/MCP, CodeQL | `agentic-security/*` (4), `quality/codeql` |
| ⚙️ `platform-engineer` | CI/CD, GitHub/GitLab PR lifecycle, worktrees, integrations, loops/swarm, triage, cost | `forge/*` (7), `integrations/*` (5), `loops/loop-runner`, `ops/swarm*`, `tooling/cli-for-agents` |
| 🔬 `researcher` | Spikes, single evidence-intake map (`project-assessment-evidence`), framework/docs exploration | `delivery/spike`, `delivery/project-assessment-evidence` |
| 🗄️ `data-engineer` | dbt/Snowflake read-only validation, notebook scaffolding (conditional — data repos only) | `data/dbt-validation`, `data/snowflake-validation`, `tooling/jupyter-notebook` |

**Orchestrator** (not in daily count): `client-workflow-bootstrap` — meta-generates `` packs/knowledge into `~/.ai-workspace` (interview-driven; not a delivery persona).

**Specialists (6 opt-in, archived 7 → `references/` #865):** `code-reviewer`, `security-reviewer` + `agentic-security-reviewer`, `e2e-runner`, `tdd-guide`, `build-error-resolver` — kept per the agent-vs-skill rule (independent verification, disjoint surface, noisy output, large diagnostic context). Archived specialists live on as inline references (`reviewer/references/`, `researcher/references/LOOKUP_GUIDE.md`, `platform-engineer/references/WORKSTATION_OPS.md`) — no knowledge deleted. Full migration map and routing chains: [`docs/AGENT_TAXONOMY.md`](docs/AGENT_TAXONOMY.md) §3/§8.

Full catalog: [`catalogs/agent-catalog.yaml`](catalogs/agent-catalog.yaml) · taxonomy: [`docs/AGENT_TAXONOMY.md`](docs/AGENT_TAXONOMY.md) · routing: [`docs/SKILL_ROUTING.md`](docs/SKILL_ROUTING.md) · personas on disk under `agents/` (see inventory)

---

## 🔄 Loop Engineering

Loops are recurring agentic workflows that run on a schedule or cadence. They follow a three-tier **mutation-safety** model enforced by `loop-gh-gate` (cadence is independent of tier):

| Tier | Mutation posture | Purpose |
|------|------------------|---------|
| **L1** | Observe / propose | Read-only or proposal-only — no repository mutations |
| **L2** | Controlled mutations | Allowlisted writes (label, comment, limited housekeeping) — merge/close denied |
| **L3** | High-autonomy mutations | Mature allowlisted mutations including merge/close when explicitly permitted |

### Loop Templates

| Template | Tier (Stage) | Default Cadence | Description |
|----------|--------------|-----------------|-------------|
| `changelog-drafter` | L1 | 1d | Draft release notes from merged PRs (L1, report-only) |
| `ci-sweeper` | L2 | 15m | Detect CI failures and propose fixes via draft PRs (L2, cautious) |
| `daily-triage` | L1 | 1d | Triage new issues and propose labels (report-only) |
| `dep-sweeper` | L2 | 1d | Apply patch-level dependency updates via draft PRs (L2) |
| `issue-triage` | L1 | 4h | Propose labels and routing for new issues (L1, propose-only) |
| `oss-daily-briefing` | L1 | 1d | Daily read-only briefing across OSS ecosystem repos (L1) |
| `oss-pr-monitor` | L3 | 1d | Monitor open PRs across OSS ecosystem repos and take action (L3, daily) |
| `oss-triage` | L1 | 1d | Triage issues across OSS ecosystem repos (L1, daily) |
| `post-merge-cleanup` | L2 | 6h | Off-peak housekeeping after merges (L2, low impact) |
| `pr-babysitter` | L2 | 15m | Mon

…

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ulises-jeremias](https://github.com/ulises-jeremias)
- **Source:** [ulises-jeremias/agent-toolkit](https://github.com/ulises-jeremias/agent-toolkit)
- **License:** MIT
- **Homepage:** https://ulises-jeremias.github.io/agent-toolkit

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-ulises-jeremias-agent-toolkit
- Seller: https://agentstack.voostack.com/s/ulises-jeremias
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.