Mycelium Adds Discovery Layer to Claude Code Agents
The open-source Mycelium repository under haabe/mycelium describes a Claude Code harness with 44 skills and 12 theory gates for discovery before delivery. The project's GitHub README frames the problem as agent-assisted teams moving too quickly from idea to code without evidence, user context, or decision provenance. For practitioners, the useful signal is not a new model but a workflow pattern: agent tooling is adding gates for product discovery, risk review, and auditability. That can reduce wasted builds, but teams still need to verify whether the framework fits their process and whether its gates are enforced consistently.
Mycelium is a small but useful signal about where coding-agent tooling is moving: away from pure code generation and toward evidence-gated delivery workflows. The practitioner value is the discipline around deciding what should be built before an agent starts producing implementation artifacts.
What happened
The open-source Mycelium project is published on GitHub under haabe/mycelium. Its README describes a Claude Code harness that makes an agent run discovery and weigh evidence before it writes code, and the project page describes a workflow with 44 skills and 12 theory gates.
Technical context
The project sits in a growing category of agent harnesses that wrap coding tools with process controls. Instead of treating the model response as the product, these systems try to preserve rationale, assumptions, risk checks, and delivery evidence. ClaudePluginHub also lists the haabe Mycelium plugin as a Claude Code plugin focused on product discovery and evidence-gated stages.
For practitioners
The pattern is most relevant for teams using agents on ambiguous product or platform work. A discovery layer can force clearer problem statements, evidence gathering, and tradeoff records before code generation. It is less useful for narrow implementation tasks where the requirement is already proven and scoped.
What to watch
Watch for adoption signals such as stars, issue activity, example projects, and integrations with existing planning or review systems. The decisive question is whether teams keep using the gates when delivery pressure rises.
Key Points
- 1Mycelium wraps Claude Code with discovery and evidence gates before an agent moves into implementation work.
- 2The value is process control for ambiguous product decisions, not a new foundation model or benchmark result.
- 3Teams should evaluate whether the gates preserve decision provenance without adding more ceremony than the project needs.
Scoring Rationale
Useful developer-tooling signal for agent-assisted software work, with primary open-source evidence. The score is slightly lower because the project is early and impact depends on adoption rather than a broad platform launch.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
