diff --git a/TODO.md b/TODO.md
index a83106f..ac4c489 100644
--- a/TODO.md
+++ b/TODO.md
@@ -1,21 +1,36 @@
-# TODO.md — go-agentic Task Queue
+# TODO.md -- go-agentic
 
-## Phase 1: Post-Split Hardening
+## Phase 1: Test Coverage
 
-- [ ] **Verify tests pass standalone** — Run `go test ./...` after split. Confirm all 5 test files pass.
-- [ ] **Review allowance model** — Current allowance tracks per-workspace quotas. Verify this aligns with the multi-tenant model in core/php-tenant.
-- [ ] **Audit client.go HTTP calls** — Check timeouts, retries, error handling for AI provider API calls.
+- [ ] Verify all 5 test files pass standalone after split (`go test ./...`)
+- [ ] Add integration test for full task lifecycle: claim -> process -> complete
+- [ ] Add edge-case tests for allowance exhaustion mid-task
 
-## Phase 2: Integration
+## Phase 2: Allowance Persistence
 
-- [ ] **Context builder improvements** — context.go builds prompts from workspace state. Add support for including CLAUDE.md, TODO.md, FINDINGS.md from the current project automatically.
-- [ ] **Multi-model allowance** — Track separate quotas per model (Opus, Sonnet, Haiku) not just per workspace.
-- [ ] **Streaming completions** — client.go should support SSE/streaming responses from AI providers.
+- [ ] MemoryStore is in-memory only -- state lost on restart
+- [ ] Add Redis backend for `AllowanceStore` interface (multi-process safe)
+- [ ] Add SQLite backend for `AllowanceStore` interface (single-node fallback)
+- [ ] Config already supports YAML -- wire backend selection into config loader
+
+## Phase 3: Multi-Agent Coordination
+
+- [ ] Currently single-agent -- only one Claude subprocess at a time
+- [ ] Add agent discovery (register capabilities, heartbeat)
+- [ ] Add task routing based on agent capabilities and current load
+- [ ] Add load balancing across multiple agent instances
+
+## Phase 4: CLI Integration
+
+- [ ] `client.go` talks to REST API -- wire into `core agent` CLI command group
+- [ ] `core agent status` -- show running agents, task queue depth, allowance remaining
+- [ ] `core agent dispatch <task>` -- submit task to queue from CLI
+- [ ] `core agent logs <task-id>` -- stream task output via WebSocket
 
 ---
 
 ## Workflow
 
 1. Virgil in core/go writes tasks here after research
-2. This repo's session picks up tasks in phase order
+2. This repo's dedicated session picks up tasks in phase order
 3. Mark `[x]` when done, note commit hash