Set up Sprinty for this project so we can plan the work, coordinate our agents, and finish with verified results. Keep our implementation plan, item ownership, verification evidence, and changelog available across context compaction and session restarts.

1. Inspect the current repository and identify my coding client. Preserve existing settings and sprint data.
2. Install the two official Sprinty skills from https://github.com/ebursztein/sprinty using:
   npx skills add ebursztein/sprinty --skill how-to-run-a-sprint --skill using-sprinty
   Choose this client's project-local skill installation. Do not install globally unless I ask.
3. Connect the Sprinty MCP server using my client's supported configuration. The server command is npx with arguments ["-y", "sprinty-mcp@latest"]. For Codex CLI use `codex mcp add sprinty -- npx -y sprinty-mcp@latest`; for Claude Code use `claude mcp add sprinty -- npx -y sprinty-mcp@latest`. For another MCP client, use its documented configuration rather than guessing. Avoid duplicating an existing server entry. Skills installation and MCP server connection are separate steps.
4. If my client needs a restart to discover the server or skills, explain the next action and stop until it is available. Do not claim the server is connected before verifying its tools.
5. Read the installed Sprinty skills. Call info({ git_dir: <canonical repository> }) first. Resume a suitable existing sprint with sprint_resume({ git_dir, data_dir }); create a sprint only when I provide work to track and no suitable sprint exists. Use a gitignored data directory, normally <repository>/.sprinty.
6. Share the dashboard URL. Create dedicated items before implementation, claim each item before working on it, record meaningful progress, and resolve it with a real commit plus executable verification evidence. Keep credentials and local sprint state out of git.
7. Each collaborating session must use the same canonical repository and data directory, with its own identity. Sprinty identifies ordinary MCP connections automatically. Claude subagents that share a connection need the official Claude plugin's identity hook, or explicit per-agent tokens as a fallback; the skills CLI alone does not install plugin hooks. Separate worktrees use separate bindings.
8. After compaction or a restart, call info and sprint_resume, then inspect next(), agent_list(), and items_claimed({ mine: true }). Recover the recorded plan and evidence rather than reconstructing progress from conversation memory. Stable native caller IDs recover an existing identity when the client supplies them; otherwise a restarted process may receive a new identity. Never assume it owns another agent's claims.

Finish setup by reporting what is installed, which tools are available, and whether I need to restart my client. Do not publish packages, deploy a website, or start unrelated work as part of setup.