Skip to content

Repository files navigation

Plasma

Plasma logo animation

Plasma is a local-first dashboard for exploring AI agent skills and reusable workflow instructions. It scans installed SKILL.md files, builds a visual skill graph, and helps you understand which skills exist, what they are for, how they relate, and which ones are worth using next.

The app is designed for people with many skills, plugins, agents, or workflow helpers installed. It supports Codex-style SKILL.md libraries out of the box and can also scan labeled SKILL.md roots for other agent environments such as Claude, Gemini, Manus, or your own instruction folders. It runs locally and does not require an account, backend, telemetry service, or cloud sync.

Plasma dashboard demo

Beta Status

Plasma is beta software. The core dashboard is usable, but the graph model, generated data shape, and activation workflows may still change as the project matures.

Features

  • Local skill inventory from Codex-style folders plus optional labeled agent roots
  • Plasma graph and glass globe views for exploring skill relationships
  • Task-first skill suggestions through the mission bar
  • Category, source, overlap, and duplicate signals
  • Local token footprint estimates for each SKILL.md
  • Optional local usage scan for supported session transcripts (Codex logs today)

GitHub Skill Check

Paste a GitHub repo or folder link into Plasma before installing a skill. The dashboard analyzes the link against your local skill graph, highlights likely overlap, and gives you the install command only after you can see what it matches.

Plasma GitHub skill check demo

Privacy

By default, npm run dev and npm run build scan installed skill files only. They do not scan agent session transcripts unless you explicitly opt in.

Optional usage scanning currently reads local Codex session JSONL files to count skill invocations. The scan is read-only and writes generated JSON to src/data/generated/, which is ignored by git.

See PRIVACY.md for details.

Quick Start

Requirements:

  • Node.js 22 or newer
  • npm
  • a local SKILL.md directory; Codex-style layouts under ~/.codex work out of the box

Install and start the local app:

Windows PowerShell:

git clone https://github.com/hkuxlicious/plasma.git
cd plasma
npm install
npm run dev

macOS / Linux:

git clone https://github.com/hkuxlicious/plasma.git
cd plasma
npm install
npm run dev

Open the local URL printed by Vite, usually:

http://127.0.0.1:5173/

Add Agent Libraries

Plasma scans CODEX_HOME or ~/.codex by default. To add Claude, Gemini, Manus, or any other SKILL.md-compatible instruction folder, set PLASMA_AGENT_SKILL_ROOTS before running npm run dev.

Use Label=path entries. Separate entries with ; on Windows and : on macOS/Linux. Adjust the paths to wherever you keep each agent's exported or converted SKILL.md folders.

Windows PowerShell:

$env:PLASMA_AGENT_SKILL_ROOTS="Claude=$env:USERPROFILE\.claude\skills;Gemini=$env:USERPROFILE\.gemini\skills;Manus=$env:USERPROFILE\.manus\skills"
npm run dev

macOS / Linux:

export PLASMA_AGENT_SKILL_ROOTS="Claude=$HOME/.claude/skills:Gemini=$HOME/.gemini/skills:Manus=$HOME/.manus/skills"
npm run dev

Include Local Usage Data

To include real usage heat, session counts, and token-footprint estimates based on reads:

npm run generate:local
npm run dev

Equivalent environment opt-in:

SKILL_DASHBOARD_SCAN_USAGE=1 npm run generate:usage

On Windows PowerShell:

$env:SKILL_DASHBOARD_SCAN_USAGE="1"
npm run generate:usage

Configuration

Plasma reads Codex-style skill directories from:

CODEX_HOME/skills
CODEX_HOME/skills/.system
CODEX_HOME/plugins/cache

If CODEX_HOME is not set, it defaults to ~/.codex.

Additional agent roots are read from:

PLASMA_AGENT_SKILL_ROOTS=Label=/path/to/skills

Generated Data

Runtime data is generated under:

src/data/generated/

This directory is ignored by git because it may contain local skill names, local paths, and optional local usage counts.

Scripts

npm run generate        # scan skills, preserve or create empty usage data
npm run generate:local  # scan skills and opt into local usage scan
npm run dev             # generate safe local data and start Vite
npm run build           # generate safe local data and build production assets
npm run serve -- 4181   # serve the built app from dist

Contributing

Contributions are welcome while Plasma is in beta. Start with CONTRIBUTING.md, keep changes focused, and run npm run build before opening a pull request.

Open Source Notes

Do not commit src/data/generated/, dist/, output/, local logs, local environment files, screenshots with private paths, or editor/agent state. The public repo should contain source code, docs, and reusable assets only.

License

Plasma is released under the MIT License.

About

Local-first visual dashboard for exploring and activating installed AI agent skills.

Topics

Resources

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages