agentjoy-bridge

agentjoy-bridge

AgentJoy Bridge is a tiny, dependency-free CLI that converts agent/tool activity into AgentJoy Run events.

It supports:

• OpenAI Codex CLI (codex exec --json)
• Claude Code hooks / plugin hooks
• (Template included) ChatGPT Apps SDK viewer (MCP)

---

Install

pipx install agentjoy-bridge
# or: pip install agentjoy-bridge

Windows (PATH): if agentjoy command is not found, run via module:

python -m agentjoy_bridge init --api http://127.0.0.1:8000 --open

補足：python が見つからない場合は py -m agentjoy_bridge ... を試してください。

---

Doctor (diagnostics)

If setup is confusing (PATH, backend, config), run:

agentjoy doctor

It prints:

• backend reachability (/health)
• config/profile status
• whether your API key / user token work

Optional (creates a small run):

agentjoy doctor --write-test

Secrets are redacted in the output.

---

Profiles (multiple environments)

You can keep multiple connection settings (local/staging/prod) and switch:

agentjoy profile set local --api http://127.0.0.1:8000 --use
agentjoy profile set staging --api https://staging.example.com --api-key ... --user-token ...

agentjoy profile list
agentjoy profile use staging
agentjoy profile show staging

For one-off commands without switching the current profile:

agentjoy demo --profile local --open
agentjoy health --profile staging

---

3-minute Quickstart (no Codex required)

Fastest: one command onboarding

If you're running from the AgentJoy repo (so ./backend exists), this command can:

• start backend (if needed)
• create a workspace + save config
• create a demo run + open the share URL

agentjoy quickstart --api http://127.0.0.1:8000 --open

Windows (PATH):

python -m agentjoy_bridge quickstart --api http://127.0.0.1:8000 --open

補足：python が見つからない場合は py -m agentjoy_bridge ... を試してください。

1) Start AgentJoy backend

cd backend
python -m uvicorn agentjoy.main:app --reload --port 8000

2) Bootstrap a Workspace + save config

This creates a workspace and stores config at ~/.agentjoy/config.json.

agentjoy init --api http://127.0.0.1:8000 --open
# (alias: --api-base)

3) Create a demo run + open the share URL

agentjoy demo --open

---

Codex (recommended)

agentjoy codex-run "Fix failing tests"

It will print a share URL you can open in a browser.

---

Claude Code

See integrations/claude-plugin/agentjoy/ for hook configs and examples.

---

Compatibility: upstream JSON changes

LLM tools evolve quickly. If a new Claude/Codex version changes the JSON field names that the Bridge reads, you can adjust the extraction paths without waiting for a package release.

1. Create a template overrides file:

agentjoy adapters --init
agentjoy adapters

2. Edit the file shown (typically ~/.agentjoy/bridge_adapters.json) and add new dot-paths.

This feature is intentionally lightweight: unknown fields are ignored, and the Bridge keeps running.

---

License

MIT