diff --git a/TASK.md b/TASK.md
deleted file mode 100644
index 0921667..0000000
--- a/TASK.md
+++ /dev/null
@@ -1,59 +0,0 @@
-# Vector Index Layer (HNSW)
-
-**Epic:** feedback-loop-v5l
-**Foundation branch:** `feature/vector-index`
-**Target:** `main`
-
-## Objective
-
-Replace brute-force O(n) embedding search with HNSW approximate nearest neighbor indexing.
-Scale floop's behavior store from 10 to millions of vectors with sub-millisecond search.
-
-## Architecture
-
-- **SQLite remains source of truth** for embeddings (BLOBs in behaviors table)
-- HNSW index is a **derived acceleration cache** at `.floop/vector.idx`
-- If index file is missing/corrupt, rebuild from SQLite on startup
-- Library: [`coder/hnsw`](https://github.com/coder/hnsw) — pure Go, zero CGO, MIT
-
-## Stack
-
-| # | Branch | Bead | Description |
-|---|--------|------|-------------|
-| 1 | `feature/vector-index/01-core` | `v5l.1` | VectorIndex interface + BruteForceIndex |
-| 2 | `feature/vector-index/02-hnsw` | `v5l.2` | HNSWIndex wrapping coder/hnsw |
-| 3 | `feature/vector-index/03-tiered` | `v5l.3` | TieredIndex auto-selector |
-| 4 | `feature/vector-index/04-wire` | `v5l.4` | Wire into MCP server + embedder |
-
-Each PR targets the previous branch. As stacks merge into `feature/vector-index`,
-remaining PRs auto-rebase onto the foundation.
-
-## Commands
-
-```bash
-# View all tasks
-bd show feedback-loop-v5l
-
-# View specific task with full implementation spec
-bd show feedback-loop-v5l.1   # Interface + BruteForce
-bd show feedback-loop-v5l.2   # HNSW
-bd show feedback-loop-v5l.3   # Tiered
-bd show feedback-loop-v5l.4   # Wire MCP
-
-# Build & test
-go build ./internal/vectorindex/...
-go test -race ./internal/vectorindex/...
-go test ./...
-```
-
-## Key Files
-
-| File | Role |
-|------|------|
-| `internal/vectorindex/index.go` | VectorIndex interface |
-| `internal/vectorindex/bruteforce.go` | Exhaustive search impl |
-| `internal/vectorindex/hnsw.go` | HNSW impl (coder/hnsw) |
-| `internal/vectorindex/tiered.go` | Auto-selector |
-| `internal/mcp/vector_retrieval.go` | Consumer (uses index.Search) |
-| `internal/mcp/server.go` | Lifecycle (init/close index) |
-| `internal/vectorsearch/embedder.go` | EmbedAndStore returns vector |