Audience: operators installing Mycelium on a new or existing project. Time to read: 5 min. Last updated: 2026-07-04.
Mycelium installs as a Claude Code plugin. Installation does not modify any
project-root files. It adds namespaced /mycelium:<skill> commands and a
.claude/ working tree the agent reads from. You can adopt it on a live
codebase without a migration step.
Prerequisite: Claude Code, installed and signed in (a Claude account or API key). Other supported agents: see install paths. Then, inside Claude Code:
/plugin marketplace add haabe/mycelium
/plugin install mycelium@haabe-mycelium
/mycelium:start # one command: setup + 10-minute discovery
/mycelium:start runs first-project setup and walks the opening discovery
round. On an empty canvas it routes you into /interview;
on a populated one it resumes via /diamond-assess.
Plugin auto-update is on by default. To update manually:
/plugin marketplace update haabe-mycelium
/reload-plugins
Agents that do not speak the Claude Code plugin spec still get framework value
through AGENTS.md orientation. The old npx degit channel is
no longer supported — on the current layout it lands an empty .claude/ with no
skills to invoke and no hooks to fire (see install paths).
If you are moving an existing legacy
.claude/ tree onto plugin form, see
the legacy→plugin migration guide and the agent-driven
/mycelium:migrate-from-legacy
skill.
- Adds: namespaced
/mycelium:<skill>commands; a.claude/working tree (canvas, diamonds, memory, harness) the agent reads and writes as it works. - Does not touch: your source files, build config, or any project-root file.
Skills are namespaced as /mycelium:<name> per Anthropic's plugin convention.