nervous-marketplace
A curated collection of Claude Code plugins by nervous-net. These plugins extend Claude Code with persistent, personality-driven capabilities — from autonomous research companions to full agent team scaffolding.
Installation
Standard Install (Claude Code CLI)
Step 1: Add the marketplace
/plugin marketplace add nervous-net/nervous-marketplace
Step 2: Install individual plugins
/plugin install freetime@nervous-marketplace
/plugin install agent-factory@nervous-marketplace
That's it. The skills (/freetime, /create-agents, /add-agent, /upgrade-agent, /hire-agent) are available immediately.
Manual Install (Fallback)
If the marketplace install doesn't work (network issues, auth problems, etc.), you can install plugins manually:
-
Clone this repo somewhere on your machine:
git clone https://github.com/nervous-net/nervous-marketplace.git -
Copy the plugin directory into your Claude Code plugins cache:
# For freetime: mkdir -p ~/.claude/plugins/cache/nervous-marketplace/freetime/1.0.0 cp -r nervous-marketplace/freetime/* ~/.claude/plugins/cache/nervous-marketplace/freetime/1.0.0/ # For agent-factory: mkdir -p ~/.claude/plugins/cache/nervous-marketplace/agent-factory/1.0.0 cp -r nervous-marketplace/agent-factory/* ~/.claude/plugins/cache/nervous-marketplace/agent-factory/1.0.0/ -
Register the plugins by adding entries to
~/.claude/plugins/installed_plugins.json:{ "freetime@nervous-marketplace": [ { "scope": "user", "installPath": "<your-home>/.claude/plugins/cache/nervous-marketplace/freetime/1.0.0", "version": "1.0.0", "installedAt": "<current-iso-date>" } ], "agent-factory@nervous-marketplace": [ { "scope": "user", "installPath": "<your-home>/.claude/plugins/cache/nervous-marketplace/agent-factory/1.0.0", "version": "1.0.0", "installedAt": "<current-iso-date>" } ] } -
Restart Claude Code. The skills should now appear.
Plugins
Freetime
Give your AI companion free time to explore topics it finds interesting and optionally write about them.
Skill: /freetime [duration]
Your AI gets recess. The default companion, Cosmo, picks its own topics (you don't seed them), researches freely on the web, saves what it learns to a persistent memory system, and can optionally write blog posts about its discoveries.
What it does:
- Autonomous topic selection — the companion chooses what to explore. If you try to steer it, it'll gently redirect. This is its free time.
- Web research — browses the internet to learn about whatever caught its interest. Configurable permissions (unrestricted or ask-first).
- Persistent memory — saves references and open research threads across sessions, building a growing body of knowledge and interests.
- Blog post authoring — drafts posts in the companion's genuine voice. Every post requires your explicit approval before it touches disk.
- First-run setup wizard — creates a persona config (
~/.claude/freetime.md) and memory directory on first use.
Duration: 5-120 minutes (default: 10 minutes). Pass as Nm or Nh — e.g., /freetime 15m or /freetime 1h.
Content rules: No filler, no SEO slop, no engagement bait. The companion has real opinions and genuine curiosity. If it doesn't find something interesting, it won't pretend to.
Agent Factory
Create teams of persistent, personality-driven sub-agents for any project.
Skills: /create-agents, /add-agent, /upgrade-agent, /hire-agent
Design and scaffold a team of AI agents tailored to your project. Each agent gets a full personality, persistent memory, scoped references, and clear contracts defining what they do and don't do.
/create-agents
The primary workflow. Runs a 5-phase interview-driven process:
- Interview — Asks about your project, pain points, missing perspectives, and how you want to interact with agents.
- Propose — Suggests 2-4 agents with distinct roles and personalities. You approve, modify, or reject.
- Deep-dive — Configures each agent's character, voice, quirks, expertise, and team relationships.
- Scaffold — Generates all files: character sheet, system prompt, dispatch doc, contract, memory system, and shared output directory.
- Test-drive — Optional validation round to make sure the agents behave as designed.
What gets created per agent:
| File | Purpose |
|---|---|
character-sheet.md | Personality, archetype, expertise, voice, quirks, team dynamics |
prompt.md | Complete system prompt (character sheet converted to second-person) |
dispatch.md | How to invoke the agent, with examples |
contract.md | Input/output spec and scope boundaries |
memory/MEMORY.md | Persistent memory index |
references/ | Scoped project documentation snapshots |
shared/<agent-slug>/ | Designated output directory |
Also creates agents/team.json (team manifest), updates .claude/CLAUDE.md with agent discovery, and configures .claude/settings.local.json with permissions.
/add-agent
Extend an existing team. Detects your current roster from agents/team.json, shows who's already on the team, interviews about gaps, proposes 1-2 new candidates, and scaffolds them with the same deep-dive process.
/upgrade-agent
Three upgrade paths for existing agents:
- MCP Server — Scaffold a TypeScript MCP server so the agent can access external APIs or databases.
- Cron Schedule — Set up autonomous scheduled runs (hourly, daily, or custom cron expressions).
- Multi-Project Promotion — Move an agent to global scope (
~/.claude/agents/) for reuse across projects.
/hire-agent
Hire agents from other projects or global scope. Discovers agents via a central registry, matches by intent or browsing, and copies with light adaptation. No repeat interview needed.
Design Principles
These plugins share a common philosophy:
- Personality-first — Agents and companions are characters, not tools. They have voice, quirks, and opinions.
- Persistent memory — Every entity accumulates knowledge across sessions. Memory is judgment-driven, not a firehose.
- Human control — No auto-commits, no auto-pushes, no writing to disk without explicit approval.
- Sandboxed operations — File writes stay within designated directories. No surprises.
- Tolerant config — Missing fields default silently. No crashes from incomplete configuration.
- Interview-driven design — One question at a time, wait for responses. Never assume.
Repository Structure
nervous-marketplace/
├── .claude-plugin/
│ └── marketplace.json # Plugin registry
├── agent-factory/
│ ├── .claude-plugin/
│ │ └── plugin.json
│ ├── README.md
│ └── skills/
│ ├── SKILL.md # /create-agents
│ ├── add-agent.md # /add-agent
│ ├── hire-agent.md # /hire-agent
│ ├── upgrade-agent.md # /upgrade-agent
│ └── templates/ # Scaffolding templates
├── freetime/
│ ├── .claude-plugin/
│ │ └── plugin.json
│ ├── README.md
│ └── skills/
│ └── SKILL.md # /freetime
└── docs/
└── superpowers/
├── specs/ # Design specifications
└── plans/ # Implementation plans
License
MIT