popcorn-harness

Instant on-the-fly harness assembler. Use /popcorn <task> to discover and combine available skills, agents, and commands.
Open source Open in the app JSON README (API)
About

Instant on-the-fly harness assembler. Use /popcorn <task> to discover and combine available skills, agents, and commands.
Details

Kind: Plugins
Topic: No topic detected
Publisher: seilk
Origin: gemini
Category: ferramentas
Version: 0.3.1
Stars: 1
Last push: 2026-04-11T12:32:02Z
Repository state: ativo
Language: Shell
Added: 2026-08-30 14:13:39
Updated: 2026-08-30 14:13:39
Origin id: seilk/popcorn-harness
README

<p align="center">
  <img src="assets/banner.svg" alt="popcorn-harness banner" width="100%">
</p>

<p align="center"><em>Instant on-the-fly harness assembler for Claude Code, Hermes, and OpenClaw.</em></p>

**Popcorn Harness** is a cross-platform plugin that discovers your available skills, agents, and commands on the fly, then pops them together into an optimized execution harness. No manual selection. No guessing. Just describe your task and let it assemble the right capabilities for you.

```
User: /popcorn Prep this Next.js app for production

🍿 Popping harness for: Prep this Next.js app for production
Platform: Claude Code (project)  |  Tier: 3 — Full Pop

Assembled harness:
  Phase 1 (parallel): security-review + e2e-testing + seo
  Phase 2 (sequential): deployment-patterns -> docker-patterns

Proceed? [y/n/adjust]
```

---

## How It Works

1. **Detects platform** — Claude Code, Hermes, or OpenClaw
2. **Discovers capabilities** — skills, agents, and commands available in your environment
3. **Applies progressive disclosure** based on task complexity:
   - **Quick Pop** — 1 capability, executes immediately
   - **Standard Pop** — 2-3 capabilities, shows plan, confirms
   - **Full Pop** — 4-5 capabilities or ambiguous task, full confirmation flow
4. **Assembles execution graph** — parallel phases where possible, sequential where required
5. **Executes and synthesizes** — structured output with key findings and action items

---

## Platform Support

| Platform | How capabilities are discovered |
|----------|--------------------------------|
| Claude Code | `claude agents` CLI + skill dirs (`~/.claude/skills`, `~/.claude/plugins/*/skills`) + command dirs + ECC marketplace (if installed) |
| Hermes | `available_skills` injected in system prompt |
| OpenClaw | `available_skills` injected in system prompt |

---

## Installation

### Claude Code — recommended

Add this repo as a marketplace, then install from it:

```
/plugin marketplace add seilk/popcorn-harness
/plugin install popcorn-harness@popcorn-harness
```

`seilk/popcorn-harness` points to the GitHub repo, which contains `.claude-plugin/marketplace.json`.
The second `popcorn-harness` is the marketplace `name` field defined in that file.

After installation, run `/reload-plugins` if commands don't appear immediately.

### Claude Code — local development / testing

```bash
claude --plugin-dir .
```

Use `/reload-plugins` after making changes. The local copy takes precedence over any installed version.

### Claude Code — manual install (fallback)

```bash
git clone https://github.com/seilk/popcorn-harness ~/.claude/plugins/popcorn-harness
```

### Hermes

```bash
cp -r skills/popcorn-harness ~/.hermes/skills/
```

The skill will appear in `available_skills` on the next session.

### OpenClaw

Add the skills directory to `external_dirs` in your Hermes config:

```yaml
# ~/.hermes/config.yaml
skills:
  external_dirs:
    - /path/to/popcorn-harness/skills
```

### Team / Custom Marketplace

If you host your own Claude Code marketplace, add popcorn-harness as a source:

```json
{
  "name": "my-marketplace",
  "plugins": [
    {
      "name": "popcorn-harness",
      "source": { "source": "url", "url": "https://github.com/seilk/popcorn-harness.git" }
    }
  ]
}
```

Or point teammates directly at this repo as a marketplace:

```
/plugin marketplace add seilk/popcorn-harness
/plugin install popcorn-harness@popcorn-harness
```

---

## Usage

### Claude Code

```
/popcorn <task description>
```

Or invoke the orchestrator agent directly from any Claude Code session:

```
Use popcorn-orchestrator to: <task description>
```

### Hermes

```
popcorn — <task description>
```

Or just describe your task and say "use available skills":

```
use your available skills to review my codebase before deployment
```

### OpenClaw

Same as Hermes — the skill is automatically available once installed.

---

## Examples

**Single-domain task (Tier 1 — Quick Pop):**
```
/popcorn run a security review on this repo
-> Pops: security-review (1 capability, executes immediately)
```

**Multi-domain task (Tier 2 — Standard Pop):**
```
/popcorn audit my portfolio and suggest rebalancing
-> Pops: obsidian-finance-vault + tossctl + market-research
-> Shows plan, waits for confirmation
```

**Complex task (Tier 3 — Full Pop):**
```
/popcorn get this app production-ready
-> Pops: security-review + e2e-testing + seo + deployment-patterns + docker-patterns
-> Shows full discovery, proposes phased execution, requires explicit confirmation
```

---

## Design Principles

- **Progressive disclosure** — reveal complexity only when needed
- **Evidence-based selection** — capabilities chosen by description match, not name alone
- **Platform-agnostic** — same skill file works on Claude Code, Hermes, and OpenClaw
- **Budget-bounded** — hard cap of 5 capabilities per harness run to preserve quality
- **Fail-transparent** — missing capabilities are reported, not silently skipped

---

## Compatibility

| Component | Minimum version |
|-----------|----------------|
| Claude Code | Latest (claude CLI with agent support) |
| Hermes | Any version with `available_skills` injection |
| OpenClaw | Any version with `external_dirs` support |
| ECC (optional) | Any version — enhances discovery if installed |

---

## Troubleshooting

**"No capabilities found"**
- Claude Code: run `claude agents` manually to verify agent discovery works
- Hermes: check `skills.external_dirs` in `~/.hermes/config.yaml`
- OpenClaw: verify the skills path is correctly set in config

**"Platform detected incorrectly"**
- Set `OPENCLAW=1` environment variable to force OpenClaw detection
- Ensure `claude` CLI is in PATH for Claude Code detection

**"Harness produced poor results"**
- Be more specific about what domains to cover (e.g., "check security, run tests, and generate a deploy checklist"). More specific tasks produce better capability matching.
- Manually specify capabilities: "/popcorn use security-review and e2e-testing to..."

---

## Contributing

Inspired by and built on patterns from:
- [ECC (Everything Claude Code)](https://github.com/anthropics/everything-claude-code) — `team-builder`, `agent-sort`, `agent-harness-construction`

PRs welcome. Please include examples for any new platform support.

---

## License

MIT