CyberBrain is a local plugin marketplace for Claude Code and Codex. It packages shared skills, commands, and agent definitions used across development workflows.
| Plugin | Contents | Status |
|---|---|---|
awesome-agent-select |
Prompted subagents for review, QA, API docs, performance, tooling, and TypeScript work | Published |
tachikoma |
Skills and commands for coordinating AI CLI tools such as Codex, Gemini CLI, OpenCode, Qwen, Cline, Aider, GitHub Copilot CLI, Kiro CLI, and Kimi CLI | Published |
brain |
Skills for epistemic audits, calculation boundaries, scientific-claim review, and whole-object responsibility | Published |
These plugins are under refactoring and are not published in the Claude Code or Codex marketplaces:
mac-econotifications
- Claude Code, for
.claude-pluginmarketplace use - Codex CLI 0.79.0 or newer, for
.codex-pluginmarketplace use jq1.8.1 or newer for local validation- Optional CLI tools used by
tachikoma: Gemini CLI, OpenCode, Qwen, Cline, Aider, GitHub Copilot CLI, Kiro CLI, Kimi CLI - Keep all CLI tools used by
tachikomaupdated to their latest available release before relying on the corresponding skill.
Clone the repository and add it as a Claude Code marketplace:
mkdir -p ~/soft
git clone https://github.com/ansatzX/CyberBrain.git ~/soft/CyberBrain
claude plugin marketplace add ~/soft/CyberBrainInstall the active plugins:
claude plugin install awesome-agent-select@CyberBrain
claude plugin install tachikoma@CyberBrain
claude plugin install brain@CyberBrainUninstall:
claude plugin uninstall awesome-agent-select@CyberBrain
claude plugin uninstall tachikoma@CyberBrain
claude plugin uninstall brain@CyberBrainClone the repository, then expose its Codex marketplace at the user-level Codex plugin root:
mkdir -p ~/soft ~/.agents/plugins/plugins
git clone https://github.com/ansatzX/CyberBrain.git ~/soft/CyberBrain
ln -sfn ~/soft/CyberBrain/.agents/plugins/marketplace.json ~/.agents/plugins/marketplace.json
ln -sfn ~/soft/CyberBrain/plugins/awesome-agent-select ~/.agents/plugins/plugins/awesome-agent-select
ln -sfn ~/soft/CyberBrain/plugins/tachikoma ~/.agents/plugins/plugins/tachikoma
ln -sfn ~/soft/CyberBrain/plugins/brain ~/.agents/plugins/plugins/brainCodex reads the marketplace from:
~/.agents/plugins/marketplace.json
Codex plugin manifests:
~/.agents/plugins/plugins/awesome-agent-select/.codex-plugin/plugin.json
~/.agents/plugins/plugins/tachikoma/.codex-plugin/plugin.json
~/.agents/plugins/plugins/brain/.codex-plugin/plugin.json
Every skill includes Codex UI metadata at:
plugins/<plugin>/skills/<skill>/agents/openai.yaml
.claude-plugin/marketplace.json # Claude Code marketplace
.agents/plugins/marketplace.json # Codex marketplace
plugins/
awesome-agent-select/
.claude-plugin/plugin.json
.codex-plugin/plugin.json
agents/
skills/
tachikoma/
.claude-plugin/plugin.json
.codex-plugin/plugin.json
commands/
skills/
brain/
.claude-plugin/plugin.json
.codex-plugin/plugin.json
skills/
mac-eco/ # disabled, under refactoring
notifications/ # disabled, under refactoring
tachikoma provides skills for running and coordinating external AI CLI tools.
Included skills:
codexgemini-cliopencodeqwencline-cliaidergithub-copilot-clikiro-clikimi-cli
Included commands:
collab-fixTDD-debug
The codex skill supports the Codex profile llm_router for routing Codex traffic through ansatzX/llm_router. llm-router adapts Codex requests and responses for non-OpenAI providers while Codex continues to execute local tools.
Supported tachikoma::codex profile policy:
| Profile | Model policy |
|---|---|
| default profile | gpt-5.4 high or gpt-5.5 high; fallback to gpt-5.3-codex when no model preference is given |
llm_router |
deepseek-v4-pro xhigh only |
aihubmix |
gpt-5.4 high or gpt-5.5 high |
Configure llm_router:
git clone https://github.com/ansatzX/llm_router.git ~/soft/llm_router
cd ~/soft/llm_router
uv sync
export DEEPSEEK_API_KEY="sk-..."
mkdir -p ~/.codex
cp llm_router.json ~/.codex/llm_router.jsonMerge the llm_router provider/profile settings from codex.config.example.toml into ~/.codex/config.toml, then start the router:
uv run llm-router serveUse it through Codex or through tachikoma::codex:
codex -p llm_router
codex exec -p llm_router -m deepseek-v4-pro --config model_reasoning_effort="xhigh" --skip-git-repo-check -brain provides audit-oriented skills for scientific, technical, and workflow reasoning.
Included skills:
using-ansatz-brainstate-machinethink-before-you-calculateepistemic-systems-auditwhole-object-responsibility
State-machine TODO:
- Add a validator that checks state nodes for completion without verification, proxy-only evidence marked verified, unresolved object drift, and open gaps before
CLAIM_READY. - Add a summarizer that rolls child agent state nodes up into a parent node without requiring concurrent writes to the same file.
- Add optional schema export for tools that want to read
.state-machine/*.mdas structured records.
awesome-agent-select provides prompted subagents.
Included agents:
api-documentercode-reviewerllm-architectmcp-developerperformance-engineerqa-experttest-automatortooling-engineertypescript-pro
Validate marketplace and plugin JSON:
jq -e . .claude-plugin/marketplace.json .agents/plugins/marketplace.json plugins/*/.claude-plugin/plugin.json plugins/*/.codex-plugin/plugin.jsonCheck skill metadata coverage:
find plugins -path '*/skills/*/SKILL.md' -type f | wc -l
find plugins -path '*/skills/*/agents/openai.yaml' -type f | wc -l

