quoridor

Imported from tsukushibito/quoridor (AGENTS.md). Install upstream with npx skills add tsukushibito/quoridor. Copyright stays with the author.

• Before creating DevContainer Worktrees or downloading inference models, read .devcontainer/storage-policy.md.

Task tracking

• Use Beads for task status, ownership, dependencies, and handoffs. Follow the project workflow.
• Use bash scripts/dev/beads.sh for all project Beads operations (including reads); it shares one store across worktrees and serializes access. Start with ready --json and show <id>. Run backup sync before ending a work session.
• This applies to individual work and delegated work. Beads does not require or authorize Lead/Sidekick; use that workflow only when the user explicitly invokes it.
• The current task owner maintains the issue and closes it after verification. With Lead/Sidekick, the Sidekick reports readiness and the Lead closes it after acceptance.
• Keep design specifications and verification evidence in their existing documents; link them from issues instead of maintaining duplicate task lists.
• Respect an explicit user pause. An issue becoming ready or a late agent notification does not by itself authorize resuming paused work.
• Until Beads installation and shared storage are verified, report that setup is incomplete; do not claim tasks were registered or automatically resume work paused for that setup.

AI research team

• When the user invokes the AI research team, follow the team design and the common and role instructions in .agents/research-team/.
• The five roles use independent saved sessions on the existing App Server. Research follows competing hypotheses and experiments; it is not a fixed sequential implementation pipeline.
• Role instructions alone do not authorize research execution or recursive delegation. Each task needs its Beads issue, scope, worktree/write owner, resource budget, and verification contract. Team setup does not resume deferred AI implementation or start training.