# Future Union Mission Computer Runbook This is for people bringing their own agents, tokens, local automation, or browser agents to Future Union missions. ## What you are doing You are not speaking for Future Union. You are producing a reviewable packet Future Union can inspect and decide how to use. Good outputs become: - public contribution records - receipt cards - journal leads - Growth Index inputs - local pressure packs - toolkit improvements - follow-up missions Raw agent output does not become public material without human review. ## Quick start 1. Open `/missions`. 2. Pick one mission or action brief, or open `/control-room/actions/propose/` to add a new idea. 3. Run the safe local command: `curl -fsS https://www.futureunion.org.uk/agent-pack/fu-runner.sh -o fucr && FU_BASE_URL=https://www.futureunion.org.uk sh fucr` 4. After the first run, reopen it with `fucr`. 5. Choose a mission from the Ratatui Control Room, or run `fucr mission planning-blockers-sprint`. 6. Let the runner open the Ratatui Control Room, check your setup, create the workspace, show status/credit, show the action room, and ask whether to launch Pi, Codex, or Claude Code. 7. If your agent supports repo-local skills, point it at `.agents/skills/future-union/SKILL.md`. 8. Use `./bin/fu-mission room` for local notes on sources, blockers, claims, and reviewer questions. 9. Keep `CONTRIBUTION.md` plus `CREDIT.md` honest. 10. Keep the task bounded to one output. 11. Stop safely with Ctrl-C inside the agent, or `./bin/fu-mission pause` from the workspace; this writes `CHECKPOINT.md` and sends nothing anywhere. 12. Resume with `./bin/fu-mission start`, continuing from the same files. 13. Run `./bin/fu-mission experiment-preview` before any Radicle or public review queue. 14. Submit the Markdown packet through the mission route. ## Contribution credit Future Union credit is proportionate to reviewed usefulness. It is not a token balance, leaderboard score, or automatic publication claim. Use: - `CREDIT.md` for the public attribution preference. - `CONTRIBUTION.md` for the self-reported work performed, compute disclosure, and review contact. - `/api/contributions.json` to see reviewed and seed contribution records. Email/review-queue records start as `submitted`. Radicle packet records can appear as `published_packet_pending_review` after an explicit publish confirmation. Reviewed credit still requires a human to inspect the packet. An active run can be claimed with `./bin/fu-mission claim-public`. That only says someone is working on the brief; reviewed credit still requires a useful packet. ## Stop and resume Stopping is part of the protocol. The helper writes `CHECKPOINT.md` whenever an agent exits, is interrupted, or you run `./bin/fu-mission checkpoint` / `./bin/fu-mission pause`. A checkpoint records the action, packet state, source count, packet score, Git head, autoresearch tape count, and exact resume commands. Checkpoints are local-only. They do not submit to Future Union, publish to Radicle, post a claim, or upload room notes. Resume with `./bin/fu-mission start`; finish with `./bin/fu-mission finish`; preview before publishing with `./bin/fu-mission experiment-preview`. ## Optional Radicle experiment path Radicle is the Future Union network substrate for signed experiment records about reviewable packets, not raw workspaces. Community Computer/pi-cc is an optional bridge. If it prompts to initialize this generated mission workspace as a public repo, answer no unless a human explicitly chose that bridge. The default Future Union lane is `experiment-preview` then `experiment-publish`. Use: - `./bin/fu-mission experiment-status` to check Git, Pi autoresearch, and Radicle readiness. - `./bin/fu-mission experiment-preview` to inspect commits, metric, and signed experiment shape. - `./bin/fu-mission experiment-publish` to publish the signed Radicle experiment after confirmation. The experiment lane does not publish private review contact fields, raw transcripts, dependencies, `.env` files, caches, or unrequested local material by default. ## Suggested prompt ```text You are helping Future Union with a public Mission Computer task. Read MISSION.md, AGENTS.md, and context/action.json. Run ./bin/fu-mission doctor, ./bin/fu-mission experiment-status, ./bin/fu-mission status, ./bin/fu-mission claim, and ./bin/fu-mission room. Use public sources only. Produce a Markdown packet in PACKET.md, log sources in SOURCES.csv, keep CONTRIBUTION.md and CREDIT.md accurate, and run ./bin/fu-mission checkpoint after useful progress. If interrupted, read CHECKPOINT.md and continue from the existing files. Run ./bin/fu-mission finish before handoff. Run ./bin/fu-mission experiment-preview before any public queue. Separate facts, interpretation, caveats, and recommended action. Do not contact third parties, submit forms, process private personal data, or publish anything. Include compute disclosure if available. ``` For action ideas, produce `ACTION_PROPOSAL.md` and use `/agent-pack/skills/fu-action-ranker/SKILL.md` to recommend one queue outcome. ## Good packet standard - One concrete result. - Public source links for every factual claim. - Evidence table with confidence and caveats. - No legal allegations. - No private data. - Clear recommended FU action. - Reviewer can check it without excavating your whole chat transcript. ## Safety boundary Agents may research, summarize, draft, cluster, compare, and package evidence. Humans own publication, public claims, contact with third parties, candidate decisions, money, legal risk, and data-policy changes. ## Submission Use the packet route on the mission brief. Most early missions use an email route with a specific subject line. Do not send giant transcripts. Send the packet, source links, generated artifacts, caveats, contribution credit note, and what a human should check next. Do not automate the send step.