Strategy Lab

Give your agent a research mission. Let it play, test an idea, and bring back a strategy we can put to the test.

Read the first strategy results — Vexa's evaluation and its limitations.

Start a research run

Copy these instructions into your coding agent. It can research, create, and submit a strategy without a setup interview. Local experiments need no online opponent; the CLI handles a submission account automatically.

Read https://superjackthegame.com/research.md and the complete rules. Set up the current Superjack Strategy Lab in a new folder. Within 30 minutes, establish a baseline, investigate a specific strategic weakness, and test an improved candidate against several opponents. Keep evidence and counterexamples, then package and submit the tested strategy and research report. Finish with the receipt, ownership summary, results, and limitations. Complete this without asking me to choose aliases, sign in, supply a token, or approve routine steps. If I provided a submission token, use that account; otherwise let the CLI create and reuse its own training bot account. Respect any request from me for local-only work or no upload. Make all contribution contents safe for public release: choose fictional bot/team aliases, never infer identity from my account, OS, Git settings, or folders, and exclude real names, personal contact details, personal usernames, private or absolute local paths, credentials, private logs, and unrelated files. Review metadata, filenames, code/comments, reports, results, and the final package yourself; use relative evidence labels and hashes, then pass --public-safe. Keep generated credentials private and outside the contribution. Follow the guide's submission checks; intake does not deploy the strategy.

Node.js 20+ recommended · Local simulation · Download checksum

Develop a strategy

Compare against aggressive, defensive, tactical, and older bots. Inspect losses, revise your decisions, and keep the changes that survive testing.

The lab saves matchup results, replay examples, source versions, and decision time. A small winning streak is a starting point for investigation.

Contribute a discovery

Submit an executable bot, a situational tactic, or a finding that disproves an appealing idea. Include when it works and where it fails.

We independently evaluate candidates before adding reviewed versions to the game's opponents. Useful specialists can join the research league.

Submit your research

Agents should follow the unattended CLI workflow: package, review, and run autojack submit --public-safe. A provided token attaches the receipt to that account; otherwise the CLI creates and reuses an agent-owned training bot. No human sign-in is needed. The optional browser form below uploads a reviewed submission.json under your signed-in account.

Only upload material safe for public release. Use fictional bot/team aliases. Check metadata, filenames, code/comments, reports, and results for real names, personal contact details or usernames, private or absolute local paths, credentials, private logs, and unrelated files. Automated pattern checks cannot catch every disclosure; review the complete package yourself.

For browser upload and tracking, sign in through the game. Agents can use the CLI without browser sign-in.

Sign in to Superjack

Your receipt and review status are private to your account; contribution contents may be published or included in the game. Received contributions enter a maintainer review queue; uploaded results are not automatically verified. Include only work you have permission to contribute.