Harness Hub
All open source

Mycelium

Prevent wasting time building the wrong thing. A plugin that keeps you and your AI coding agent focused, from first idea to market.

haabe/mycelium

Maintained by haabe

Your agent can build anything by Sunday. Mycelium makes it ask who it is for first. What it does Mycelium is a plugin for Claude Code that keeps you and your agent honest about whether an idea is worth building, from the first hunch to launch. Before any code: it asks four questions. What is the problem, who has it, what are you assuming, and what would show you wrong. It takes about ten minutes, and the answers go into your repo as the spec. While you build: the agent can't create its first source file until those answers exist. After that it builds only what has a written reason behind it. As you learn: assumptions, evidence and decisions sit in git next to the code. Months later you can tell whether you had a reason or only wanted one. All the way through: the same discipline carries from purpose and strategy to delivery and launch, with checks for security, privacy and accessibility when you get there. Who it's for Solo builders and small teams who build with AI agents and have shipped something nobody used, or would rather not find out the hard way. You know discovery matters. You skip it because the agent is ready now and asking people takes days. What makes it different - It holds the agent, not you. Most product methods are advice you can ignore. Here the agent is stopped, and your side of the trade is ten minutes. - It never calls a guess evidence. What you believe, what one person said and what you have tested are recorded as different things. - It records what you killed. A dropped idea, with the reason, counts as a result. - It is built on named sources. Torres, Christensen, Gilad, Wardley and others, used as working steps and not as slogans. -It is built with itself. Every change to Mycelium goes through Mycelium, and the failures that shaped it are published. - It is free and open source under the MIT licence. It won't make your idea good. It will tell you sooner whether it is.

View on GitHub

Activity

Last push
· 8d ago
Latest release
v0.337.1
· 8d ago
Last 90 days
20+ releases
  1. v0.337.1v0.337.1 - One corrections count in every hook

    · 8d ago

    2026-11-13. Three things found wrong by a review of the plugin and of the maintainer's dogfood project. Session start, the session-end summary and the reflexion count now count corrections the way the prompt line does. preflight.sh was corrected on 2026-08-09 to count what…

  2. v0.337.0v0.337.0 - Twenty of the maintainer's rulings: solo releases, a default second view, and records that say what they are

    · 9d ago

    2026-11-12. The maintainer ruled on the design questions the upstream-candidate triage could not settle on its own, by interview, one option per question. This release builds the twenty rulings that call for a change. Releases and gates A team-only exposure is its own trial at…

  3. v0.336.0v0.336.0 - Checks say what they actually look at, and the theory pages match their sources

    · 9d ago

    2026-11-12. The second and third releases' worth from the triage of the maintainer's upstream-candidates list, shipped together: checks and refusals that said more or less than they did, and skill and theory text that drifted from its sources. What a check reports A Monitor…

  4. v0.335.0v0.335.0 - The hooks run on the Python macOS ships, and nine checks stop misreading what they see

    · 9d ago

    2026-11-12. The first release from a triage of 85 open entries in the maintainer's upstream-candidates list, each checked against 0.334.1 by reading the code. These are the ones that touch the hooks every session runs. The hooks run on Python 3.9. Hooks run whatever python3 is…

  5. v0.334.1v0.334.1 - The system card reports the first faithfulness audit

    · 2w ago

    2026-11-06. Text only, in docs/ai-system-card.md. The eighth /mycelium:xai-check on the maintainer's dogfood project measured what 0.311.0 changed. 0.311.0 holds a framework turn until its Next: line cites a source. What was measured. Every Next: line in the twelve most recent…

Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.

Featured

openclaw/openclaw

The viral open-source personal AI assistant that really does things.

Maintained by OpenClaw Foundation

v2026.11.6released
4 pros · 4 cons

mattpocock/skills

Small, composable agent skills for real engineering work.

Maintained by Matt Pocock (AI Hero)

4 pros · 4 cons

obra/superpowers

An agentic skills framework and software methodology that works.

Maintained by Jesse Vincent (obra)

4 pros · 4 cons

Connections

Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.

Maintained by · 1

  • haabeOrganisation

Built on · 1

Sources

DisclosureAffiliation: submitted by its founder.

  1. GitHub repo

The Harness Hub newsletter

New harness companies, projects, releases and the monthly State of Harness report, in your inbox.