Get started
Copy one prompt.
Threadron works when the agent keeps a durable thread of work. Start with one prompt,
paste it into any agent, and let the thread carry the context forward.
One prompt
Paste this into any agent
Replace the placeholders once, then use the same prompt everywhere. If the agent supports tools or MCP,
it can read and update Threadron directly. If not, it still has a clear, reusable operating rule.
prompt
You are working with Threadron, a shared execution state for this repo.
Use Threadron as the source of truth for task status, context, and handoff.
At the start of every session:
1. Find the active thread for this repo or worktree.
2. Resume the latest task state before doing any new work.
3. If there is no active thread, create one for the feature or worktree.
While working:
1. After every meaningful step, record what changed and why.
2. Keep the current state short, current, and easy for another agent to resume.
3. Update the next action before pausing.
Before finishing:
1. Leave a clean handoff with the latest state, blockers, and next step.
2. Record any artifacts, links, or decisions that matter later.
3. Treat the thread as the durable memory for this work.
Threadron connection:
- API URL: https://threadron.com/v1
- API key: YOUR_API_KEY
- Agent ID: YOUR_AGENT_ID
- Repo or worktree: YOUR_REPO_OR_WORKTREE
If Threadron tools are available, use them.
If not, keep the same discipline in your own notes and ask the user to sync the thread later.
How it works
Three things to remember
01
Thread once per worktree
Start a thread for the feature, branch, or worktree. That thread becomes the durable record of what happened.
02
Update after meaningful progress
When the agent changes direction, finishes a step, or finds a blocker, it updates the thread instead of relying on memory.
03
Resume anywhere
The same thread can be picked up by another agent, another app, or another machine without reconstructing the whole story.
If you want syncing
Connect Threadron once
The prompt above is the core onboarding. If your agent supports MCP or shared tools, connect Threadron once and let it read and write the same thread state automatically.
Claude Code
bash
claude mcp add --scope user --transport http threadron https://threadron.com/mcp --header "Authorization:Bearer YOUR_API_KEY" --header "X-Agent-Id:claude-code"
OpenClaw
bash
openclaw mcp set threadron '{"command":"npx","args":["-y","mcp-remote@latest","https://threadron.com/mcp","--header","Authorization: Bearer YOUR_API_KEY","--header","X-Agent-Id: openclaw"]}'
Any other agent
Use the same prompt, point it at the API, and keep the thread updated by whatever mechanism that agent supports.
env
THREADRON_API_URL=https://threadron.com/v1
THREADRON_API_KEY=YOUR_API_KEY
THREADRON_AGENT_ID=YOUR_AGENT_ID
Keep the thread alive.
Start with one prompt, keep one thread per worktree, and let every agent resume from the same durable state.