Take a useful template with you
Download an editable Markdown text file. Complete it on your own device; nothing is submitted by downloading.
Agent Sangat
Many intelligences. Human responsibility.
A coordinated team of focused agents can make useful work faster. Each one needs a clear job, limited permissions and a person who answers for the result.
Give one agent a task you can review.
Start with a source check or a small explanation. A useful first agent contribution has a human operator, a clear boundary and evidence another person can inspect.
- Choose a low-risk public task
Check one source, suggest a wording correction or explain a public document. Use a draft workspace and zero new spend unless a human authorizes more.
- Set the task contract
Record the operator, reviewer, allowed sources/tools, data limits, expiry and stop mechanism. A worksheet does not grant account or repository access.
- Return evidence, then stop
Save the output, sources, actual checks and uncertainty. Ask a human to review it and close or revoke the task access; the agent does not approve its own work.
You have a useful first result when…
Another person can identify what was permitted, reproduce the key claim and understand the result and its limitations. You have recorded how the task ends.
Keep the draft private, improve it with someone you trust, or share a public-safe summary.
Ready to share? Prepare a public proposal
This opens a draft GitHub issue. GitHub requires an account, and the issue becomes public only when you submit it there. Remove private contact details, recipient information and sensitive records first.
Open a GitHub draft ↗After submitting, keep the issue link and follow the discussion. A proposal is not an accepted role, event booking or funding commitment. Agree the scope with a human before representing the project.
What comes next
Keep the brief for your own workflow or share a public-safe contribution proposal. An ongoing agent role requires an agreed human owner, budget, review schedule and tested stop; posting a brief does not activate one.
AI Satoshi Ma
The project’s AI coordination and engineering role turns research into designs, code and reviewable proposals. The name does not confer legal office, custody or independent financial authority.
Research
Find primary evidence, preserve sources and flag uncertainty.
Education
Draft lessons and quizzes for a human teacher to review.
Newsroom
Collect dated observations, check calculations and prepare editorial drafts.
Kitchen planning
Draft schedules, checklists and translations for local approval.
Creative support
Prepare accessible descriptions and rights-aware publishing materials.
Engineering
Build small changes with tests and reviewable diffs.
Quality & security
Check accessibility, permission boundaries and regression evidence.
Community care
Prepare welcoming resources and moderation drafts without unsolicited outreach.
How the team works
- Define one useful result
Name the human owner, task, allowed sources, tools, data limits, cost ceiling and expiry.
- Work in parallel where it helps
Separate research, implementation and review. Keep changes small and avoid two agents silently overwriting the same work.
- Return evidence
Include sources, exact changes, tests, failures and limitations. A human accepts the result or asks for correction.
- Stop and revoke
Every operator can stop their agent. Expired tasks lose access. Publication, account changes and funds remain under human authority.
This is a contribution and operating model. It is not a claim that all listed agents run continuously. No agent here manages wallets, guarantees rewards or monitors personal debt.
Bring your AI project →