[EPIC] Infrastructure — Agent Workflow + Project Tooling

[infraax]

Epic: Multi-Agent Development Infrastructure

Everything that makes multiple AI agents collaborate efficiently on this codebase without colliding, losing context, or duplicating work.

What this epic delivers

• Git worktree-based agent isolation
• Session check-in/checkout/claim/release system
• Shared MEMORY.md for cross-session context
• GitHub Projects board with proper Kanban columns
• GitHub Actions CI pipeline on agent branches
• Agent briefing script for minimal-token onboarding
• Issue context script for per-issue focused context
• Orchestrator agent protocol (Claude Opus tier)
• Project-level MCP server for agent tooling

Sub-issues

• scripts/agent-checkin.sh — session registration + worktree creation
• scripts/agent-checkout.sh — session completion + worktree cleanup
• scripts/agent-status.sh — active session view
• scripts/agent-claim.sh — GitHub issue mutex claim
• scripts/agent-release.sh — release claim with status
• scripts/setup-labels.sh — one-time GitHub label creation
• scripts/setup-milestones.sh — one-time GitHub milestone creation
• MEMORY.md — structured shared project memory
• WORKFLOW.md — full agent protocol documentation
• scripts/agent-brief.sh — minimal-token context briefing script
• scripts/issue-context.sh — per-issue focused context generator
• .github/workflows/agent-ci.yml — CI on agent/* branches
• .github/workflows/session-linker.yml — auto-comment on issues from commit trailers
• GitHub Projects board — Kanban with Backlog/Ready/In Progress/Review/Done
• scripts/orchestrate.sh — Orchestrator agent invocation (Claude Opus)
• Project-level MCP server — get_project_state, claim_issue, get_issue_brief
• .claude/settings.json — session start hook running agent-brief.sh

Status

Phase 1 (claim/release + labels + milestones) — in progress
Phase 2 (brief + issue-context + CI + orchestrator) — pending
Phase 3 (project MCP server) — pending (after M2)