A Claude Code plugin bundling the Fused _core management skills — the ones that read and write the Fused App state store (tasks, runs, feedback cards, secrets, the agent roster, artifact chats).
These skills drive the _core workspace UDFs — the App state store at ~/.openfused/app/state.json — over the local execution layer started with fused dev serve. Reads go through SQL (/api/exec/sql), writes through UDFs (/api/exec/udf); each skill documents its own access pattern and is usable on its own.
Driving the
fusedCLI for setup, running code, and building projects/widgets lives in theagent-coreplugin infusedio/skills. This repo is management-state only.
Load the repo as a plugin:
claude --plugin-dir /path/to/flow-skillsThe manifest at .claude-plugin/plugin.json points at the skills/ directory ("skills": "./skills"), so every skill below loads at once. Each skill is also self-contained and usable on its own.
For the CLI-driving usage skills (setup, infra, projects, execute, verify, storage, widgets, feedback), also load the companion agent-core plugin alongside this one:
git clone https://github.com/fusedio/skills
claude --plugin-dir /path/to/flow-skills --plugin-dir /path/to/skills/agent-core| Skill | Purpose |
|---|---|
task-management |
Read, create, assign, and re-status tasks and their comments; render the standalone task-board widget. |
run-management |
Read and write agent run records and per-run transcripts. |
feedback-management |
Read and write interaction cards — the system of record for HITL decisions. |
secrets-management |
Get, put, list, and delete secrets in the local Fernet-encrypted store. |
agents-management |
Create, read, update, delete, clone, and reset agent-roster entries. |
artifact-chat-management |
Create and update artifact chat threads and their messages/transcripts. |
Each skill is a Fused _core project: a set of UDFs under scripts/<op>/ (each with a main.py + spec.md), an openfused.toml that registers them, and — where relevant — a shipped widget under widgets/. See any skill's SKILL.md for its operations and the read/write access pattern.
Two kinds of change, handled in opposite ways — see CONTRIBUTING.md:
- Changing what a skill means (new defaults, a repurposed op, a team-specific workflow) → don't edit the shipped skill. Create a new skill or Fused project in your own workspace and diverge there.
- Fixing a bug or adding to a skill (doc fix, wrong/missing field, an additive op) → contribute it back as a PR so the fix lands upstream instead of living as a local divergence.