CodexPluginStats
← Claude plugins
CLAUDE PLUGIN

larch

Sergey Zhupanov v56.2.2

Larch is a multi-agent workflow automation plugin for Claude Code that orchestrates the full software development lifecycle — from design through merge — using collaborative AI agents. It provides ...

Create a free account to save this plugin.

From the developer

License

MIT

Works with

Claude Code

Description

From the marketplace listing

Larch is a multi-agent workflow automation plugin for Claude Code that orchestrates the full software development lifecycle — from design through merge — using collaborative AI agents. It provides six slash commands (/design, /implement, /review, /research, /loop-review, /alias) that coordinate up to 5 parallel agents (Claude subagents + Codex + Cursor) with a voting-based review system to adjudicate findings. Key features include: diverge-then-converge design planning where independent agents propose architectures before anchoring bias can set in; multi-round code review with a 3-agent voting panel (YES/NO/EXONERATE) that accepts or rejects each finding; end-to-end automation from feature design through PR creation, CI monitoring, Slack announcements, and merge; a reviewer competition scoring system that tracks finding quality; and systematic whole-repo review via partitioned slices. External tools (Codex, Cursor, Slack) are optional — the plugin degrades gracefully when they're unavailable.

Source repository ↗ Website ↗

What changed

No changes recorded yet. Follow this plugin to keep track of future changes.

Keywords

Files & skills

larch · catalog manifest

Saved declaration

{}
larch · catalog manifest

Saved declaration

{
  "name": "larch",
  "source": {
    "sha": "1d563631f0365174d2653c12cda626c008e5ec03",
    "url": "https://github.com/zhupanov/larch.git",
    "source": "url"
  },
  "homepage": "https://github.com/zhupanov/larch",
  "description": "Larch is a multi-agent workflow automation plugin for Claude Code that orchestrates the full software development lifecycle — from design through merge — using collaborative AI agents. It provides six slash commands (/design, /implement, /review, /research, /loop-review, /alias) that coordinate up to 5 parallel agents (Claude subagents + Codex + Cursor) with a voting-based review system to adjudicate findings. Key features include: diverge-then-converge design planning where independent agents propose architectures before anchoring bias can set in; multi-round code review with a 3-agent voting panel (YES/NO/EXONERATE) that accepts or rejects each finding; end-to-end automation from feature design through PR creation, CI monitoring, Slack announcements, and merge; a reviewer competition scoring system that tracks finding quality; and systematic whole-repo review via partitioned slices. External tools (Codex, Cursor, Slack) are optional — the plugin degrades gracefully when they're unavailable."
}
larch · catalog manifest

Saved declaration

{
  "name": "larch",
  "author": {
    "name": "Sergey Zhupanov",
    "email": "zhupanov@yahoo.com"
  },
  "license": "MIT",
  "version": "56.2.2",
  "homepage": "https://github.com/character-ai/larch",
  "keywords": [
    "workflow",
    "automation",
    "code-review",
    "ci",
    "pull-request",
    "multi-agent",
    "design-review",
    "pr-automation",
    "collaborative-review",
    "voting-panel",
    "issue-dependencies"
  ],
  "repository": "https://github.com/character-ai/larch",
  "userConfig": {
    "codex_model": {
      "type": "string",
      "title": "Codex Model",
      "sensitive": false,
      "description": "Strong/default Codex model for untagged launches such as brainstorm, negotiations, health probes, and callers that do not pass --model-role. Also accepted via LARCH_CODEX_MODEL. When unset, defaults to gpt-5.6-sol. Step 2 Codex implementers are difficulty-routed unless pinned: TRIVIAL uses gpt-5.6-terra, MODERATE/HARD use gpt-5.6-sol. Review panels, voters, and fixers use their role env vars and tier routing before falling back to gpt-5.6-luna or gpt-5.6-terra."
    },
    "codex_effort": {
      "type": "string",
      "title": "Codex Effort",
      "sensitive": false,
      "description": "Codex reasoning effort at launch sites that pass --with-effort or use internal effort-enabled launchers (review, implementation, voting, health probe): minimal, low, medium, or high. Unlike model routing, there are no per-role effort variants—review, vote, and fix roles share this same effort level. Also accepted via LARCH_CODEX_EFFORT env var. When unset, defaults to 'high' at those launch sites. launch-codex-exec callers that omit --with-effort (design drafter, /research lanes, plan autofix, lint-fix) ignore this setting."
    },
    "cursor_model": {
      "type": "string",
      "title": "Cursor Model",
      "sensitive": false,
      "description": "Default model for Cursor launches (e.g., gpt-5.4-medium). Also accepted via LARCH_CURSOR_MODEL env var. Reviewer-panel rows may override this per slot with auto. When unset, defaults to composer-2.5 (with max-mode engaged via scripts/cursor-wrap-prompt.sh)."
    }
  },
  "description": "Multi-agent workflow automation for Claude Code: issue-anchored `/design` (one direct-drafting flow; plan-review panel), `/bug` (consumer issue-filing helper), `/deps` issue dependency audits, and `/implement` (positional `<issue-N>`, fixed 7200s Step 2 timeout, optional `--merge`, `--forked`, `--draft`, `--no-admin-fallback`, `--no-logs-commit`, `--coder`, `--run-id`). `/implement` Preflight reads the `larch:plan` block from the issue body; `/implement` Step 5 code review always uses the unified strict panel internally (public `--panel` argv removed)."
}

plugin.json observed Nov 6, 2026 · 10:45 UTC. Manifest source ↗

Catalog declarations and available plugin.json metadata are saved. Referenced skills and executable source files are not downloaded or tested. Missing components in a manifest do not prove that the plugin has none.

Installation instructions

Community

/plugin marketplace add anthropics/claude-plugins-community
/plugin install larch@claude-community

Claude Code installation documentation

Technical details
Sources
Community
First seen
Oct 3, 2026 · 22:13 UTC
Last seen
Nov 7, 2026 · 12:00 UTC
Latest observed change
Oct 3, 2026 · 22:13 UTC
Collection status
Listed
Reported installs
Not provided

Installation counts are source-reported, not independently verified active users. Observation dates are not release dates.

Download plugin data (JSON)

Plugin history: served from cache. Prepared . Refreshed when catalog data changes or after 6 hours. Your account and Follow status are live.