Anthropic Claude Plugin
The claude plugin integrates Anthropic Claude Code CLI (claude) and Claude Desktop into a Repo Nexus Virtual Meta-Repo.
All AI assistants in Repo Nexus share [AGENTS.md](file:///Users/admin/Projects/ai/repo-nexus/AGENTS.md) at the workspace root as the single universal source of truth for routing. The claude plugin mirrors human-facing operational prompts and plugin workflows into .claude/commands/ and .claude/prompts/ so they can be triggered directly as custom slash commands (e.g. /rnex-cross-repo-feature, /rnex-workspace-audit, /rnex-wiki-ingest).
Operational skills, workflows, rules, and subagents remain cleanly self-contained within .rnex/ and active plugin directories (.rnex/plugins/<plugin-name>/).
Capabilities
Custom Slash Commands (
.claude/commands/*.md&.claude/prompts/*.md):- Mirrors standardized Repo Nexus operational workflows into
.claude/commands/*.mdand.claude/prompts/*.md. - In Claude Code CLI, workflows can be invoked directly as custom slash commands (e.g.
/rnex-cross-repo-feature,/rnex-workspace-audit,/rnex-wiki-ingest).
- Mirrors standardized Repo Nexus operational workflows into
Native Instruction Support via
CLAUDE.md&AGENTS.md:- Claude Code natively reads
CLAUDE.mdas its primary instructions file; it only falls back toAGENTS.mdwhen noCLAUDE.mdexists in the project. - When enabled (
instructions: true), the Claude plugin synchronizes Repo Nexus routing instructions toCLAUDE.md(or merges them cleanly if you already have custom instructions). - If
CLAUDE.mdis absent andinstructions: false, Claude Code falls back to readingAGENTS.mddirectly.
- Claude Code natively reads
Subagent & Skill Execution:
- Skills reside under
.rnex/skills/and.rnex/plugins/<plugin>/skills/. - Workflows reside under
.rnex/workflows/and.rnex/plugins/<plugin>/workflows/. - Subagents reside under
.rnex/agents/and.rnex/plugins/<plugin>/agents/and are invoked as subagents without external mirroring.
- Skills reside under
Enabling the Plugin
Via CLI
# Enable across the workspace in rnex.yaml
rnex plugin enable claude
# Enable locally in .local.rnex.yaml for your workstation only
rnex plugin enable --local claudeOr initialize a new workspace with Claude Code integration pre-configured:
rnex init --claudeConfiguration Options
Configure options under plugins.claude in rnex.yaml or .local.rnex.yaml:
plugins:
claude:
instructions: true # Sync routing instructions to CLAUDE.md (default: true)
prompts: true # Mirror prompts to .claude/commands/ and .claude/prompts/ (default: true)
workflows: true # Mirror active plugin workflows (default: true)
enabled: true # Set to false to disable| Option | Type | Default | Description |
|---|---|---|---|
instructions | boolean | true | Merge routing instructions into CLAUDE.md. |
prompts | boolean | true | Mirror prompts to .claude/commands/*.md and .claude/prompts/*.md. |
workflows | boolean | true | Mirror plugin workflows as Claude slash commands. |
enabled | boolean | true | Enable or disable the plugin. Overridable in .local.rnex.yaml. |
Generated Directory Structure
When claude is enabled with functional plugins like karpathy-llm, it generates:
my-workspace/
├── AGENTS.md # Universal workspace instructions & routing
├── CLAUDE.md # Claude-specific instructions (if instructions: true)
├── .claude/
│ ├── commands/ # Slash command shortcuts
│ │ ├── rnex-cross-repo-feature.md
│ │ ├── rnex-workspace-audit.md
│ │ ├── rnex-wiki-ingest.md
│ │ └── rnex-wiki-lint.md
│ └── prompts/ # Prompts catalog
├── .rnex/
│ ├── skills/ # Workspace operational skills
│ ├── workflows/ # Workspace workflows
│ └── plugins/
│ └── karpathy-llm/ # Scoped plugin skills, workflows, and agents
└── rnex.yamlDisabling & Pruning
To disable the Claude plugin:
rnex plugin disable claudeRepo Nexus automatically performs surgical cleanup:
- Removes auto-generated commands and prompts under
.claude/. - Cleans up any legacy mirrored skills or instructions blocks.
- Developer-created custom commands, skills, or agents under
.claude/are preserved.
