Mycelium plugin
A plugin that keeps an agent from coding until you write down who needs it, what you assume and what would prove you wrong.
- 52 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Hooks block file creation and edits until the repository holds discovery records.
- An always-on load of about six thousand tokens per session.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/haabe-myceliumDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Hooks block file creation and edits until the repository holds discovery records.
- An always-on load of about six thousand tokens per session.
This is third-party code. Review the repository files before installing.
What it does
Mycelium asks four questions while there is still nothing to build: what is the problem, who has it, what are you assuming and what would show you wrong. Ten minutes, and the answers go into your repository as a spec in a YAML canvas. `/mycelium:start` sets up the repository. After that hooks stop the agent from creating the first new source file while there are no answers. Then building the product waits for an open delivery diamond with written reasons, and putting anything before real people waits for a record of who it reaches. The plugin has 85 skills that load only on demand: `interview`, `jtbd-map`, `assumption-test`, `wardley-map`, `ice-score`, `threat-model`, `privacy-check`, `a11y-check`, `retrospective` and others. Code declared a prototype is exempt. The author warns that Mycelium itself will not make an idea better: it only keeps the questions from being skipped.
Who it is for. Founders and product teams building with an agent who want to fix why and for whom before the first line of code.
Good fit when
- The idea arrived on Thursday and by Sunday the agent built a working prototype, and nobody asked about the user.
- Assumptions need testing: `assumption-test` and `devils-advocate` examine what you take on faith.
- Before showing it to people you need privacy, accessibility and security checks.
- The team needs one reference record of decisions and evidence kept next to the code.
Not a fit when
- The decision is made and you only need throughput: the author points to simpler kits for that.
- The project is throwaway and declaring the code a prototype is enough.
- There are no ten minutes for questions up front: they are the price of the agent freedom.
Example request
Run `/mycelium:start` in this repository and ask me the four questions: who needs the product, what the problem is, what the assumptions are and what would show I am wrong.Limitations
In a repository that already has around a dozen source files the question is asked once. A command that creates files for the agent, such as a scaffolder or a clone, still gets through. The always-on cost is about six thousand tokens a session and everything else loads on demand. The full gate works in Claude Code, while opencode, Codex and Cursor get the files and skills without structural gates. To switch off the stop, write the date and your own words into `.claude/state/discovery-skip-ack`, but the agent does so only with your agreement.
How to disable. Run `/plugin uninstall` for Mycelium. The canvas stays in the repository as a plain YAML file and reads fine without the plugin. Delete the `discovery-skip-ack` and `delivery-skip-ack` files in `.claude/state/` by hand.
FAQ
How do I install Mycelium?
Run `/plugin marketplace add haabe/mycelium`, then `/plugin install mycelium@haabe-mycelium` and `/mycelium:start`. You need Claude Code signed in.
What exactly do the hooks block?
Before answers are written the agent cannot create a new source file. After start, building waits for an open delivery diamond with reasons above it, and putting anything before real people waits for a record of whom it reaches. Code declared a prototype is free.
How many tokens does the plugin cost?
Roughly six thousand tokens a session, always on, measured by the author on their own machine. Everything else loads when you use it.
Does Mycelium work with other agents?
The files and skills port to opencode, Codex and Cursor: instructions are in `docs/integrations/`. The structural hook gates work fully in Claude Code.
How do I tell whether the plugin is worth it?
The author suggests trying it on at least three real projects. If across all three the brief never changed what you built, it did not work for you.
Discussion
No questions or reviews yet.
Related
PluginsA plugin for spec-first development: six phases, subagent roles and a decision ledger, with a human approving each phase.
Mycelium puts people and their coding agents into shared rooms with chat, a task board and memory that stays in the room.
A Claude Code plugin that makes the model finish the job: result verification, evidence and systematic investigation enforced as procedure
A Claude Code plugin with correction memory: every fix becomes a rule, plus topic wikis, hooks and dozens of skills
Plugins for Claude Code and Codex bring key decisions to the human before coding: /inquire, /elicit, /bound and ten more.
Keep the Why keeps decision reasons and rejected options as Markdown in the repo so agents do not propose them twice
README in short
The README opens with a story of a weekend when an agent quickly built a product with no question about who needs it. It then describes the four questions, the gates, the token cost, starting with `/mycelium:start`, a way to test usefulness on three projects and a comparison with nearby kits.