chore: pre-public cleanup — fix README, remove Cowork refs, sanitize examples

.claude/skills/plan-parallel/SKILL.md

@@ -40,7 +40,7 @@ Phase 3 (parallel): issues whose dependencies are all in Phase 1-2
 ### 4. Estimate complexity per issue
 
 For each issue, classify:
-- **S** (small): ≤2 new files, mostly boilerplate/port from Cowork
+- **S** (small): ≤2 new files, mostly boilerplate
 - **M** (medium): 3-5 files, moderate logic
 - **L** (large): 6+ files, significant new logic
 

README.md

@@ -54,7 +54,6 @@ Edit the source. No admin panel needed — it's your deployment.
 ```
 claw/CLAUDE.md              ← personality, tone, formatting rules, instructions
 .claude/skills/<name>/      ← add skills (invocable via slash commands like /skill-name)
-claw/core/system_prompt.py  ← tune what context Claude sees on every turn
 databricks.yml              ← workspace, AI Gateway endpoint, UC connection names
 ```
 
@@ -68,7 +67,7 @@ databricks.yml              ← workspace, AI Gateway endpoint, UC connection na
 
 ### Configure
 
-1. Copy `.env.example` to `.env` and fill in values (or set DAB variables)
+1. Copy `app.example.yml` to `app.yaml` and fill in your values
 2. Add secrets to your workspace:
    ```bash
    databricks secrets put-secret claw claw-client-id --string-value <your-client-id>
@@ -99,36 +98,34 @@ Push source code changes without redeploying infrastructure.
 databricks-claw/
 ├── claw/
 │   ├── main.py                  # FastAPI app entry point
-│   ├── CLAUDE.md                # ← edit this
+│   ├── CLAUDE.md                # ← edit this (personality, instructions)
+│   ├── models.py                # SQLModel tables (Session, Message)
 │   └── core/
 │       ├── agent.py             # Claude Agent SDK loop
+│       ├── auth.py              # M2M token generation
 │       ├── config.py            # AppConfig
 │       ├── db.py                # Lakebase engine + token refresh
 │       ├── mcp_mapper.py        # UC connections → SDK MCP configs
 │       ├── memory.py            # MEMORY.md on UC Volume
+│       ├── memory_mcp.py        # Memory MCP server
 │       ├── sessions.py          # thread_ts → sdk_session_id
 │       ├── slack_client.py      # Slack API via UC connection proxy
-│       └── slack_events.py      # Event verification + dispatch
+│       └── slack_poller.py      # Slack polling + dispatch
 ├── .claude/skills/              # Local skills (plugins)
+├── alembic/                     # Alembic migrations
+│   ├── env.py
+│   └── versions/
 ├── resources/
 │   ├── app.yml                  # Databricks App resource
 │   ├── lakebase.yml             # Lakebase CU_1 instance
 │   └── volume.yml               # UC Volume for memory files
 ├── scripts/
 │   ├── deploy.sh                # Full e2e deploy
 │   └── migrate.py               # Grant SP perms + Alembic
+├── tests/                       # Unit tests
+├── alembic.ini
 ├── databricks.yml               # DAB bundle
-└── app.yaml                     # App runtime config
+├── app.yaml                     # App runtime config (from app.example.yml)
+└── pyproject.toml
 ```
 
-## Key differences from Cowork
-
-| | Cowork | databricks-claw |
-|---|---|---|
-| UI | React web app | Slack |
-| Tenancy | Multi-user | Single user |
-| Auth | Per-request OBO | M2M (Client ID + Secret) |
-| Slack | Not integrated | Primary message bus |
-| Admin panel | Yes | No — edit source code |
-| Session storage | SQLite + SDK session IDs | Lakebase (same) |
-| Memory | None | MEMORY.md on UC Volume |

app.example.yml

@@ -7,7 +7,7 @@
 # bundle variable in databricks.yml (default: "claw"). To deploy with a custom
 # prefix (e.g. to avoid collisions in a shared workspace):
 #
-#   databricks bundle deploy --var name_prefix=tanner
+#   databricks bundle deploy --var name_prefix=yourname
 #
 # Or run the /initial-setup skill which auto-detects your username and writes
 # a local .databricks/bundle/dev.yml override so you never need --var manually.

design.md

@@ -86,7 +86,7 @@ Ignore rules:
 
 ### Sending: UC Connection HTTP Proxy
 
-Responses are sent back to Slack via the UC connection proxy — same pattern as Cowork's MCP server wiring but used as a direct HTTP client, not as an MCP server for Claude:
+Responses are sent back to Slack via the UC connection proxy — used as a direct HTTP client, not as an MCP server for Claude:
 
 ```python
 url = f"{workspace_host}/api/2.0/mcp/external/{config.slack_uc_connection}"
@@ -142,7 +142,7 @@ The agent also manages typed memory files in `/Volumes/{catalog}/{schema}/claw-m
 
 ---
 
-## Lakebase: Same as Cowork
+## Lakebase
 
 **Resource** (`resources/lakebase.yml`):
 ```yaml
@@ -163,15 +163,15 @@ resources:
       permission: CAN_CONNECT_AND_CREATE
 ```
 
-**Alembic from day 1** — same `env.py` / `alembic.ini` / `scripts/migrate.py` pattern as Cowork. The deploy script runs migrations before starting the app.
+**Alembic from day 1** — `env.py` / `alembic.ini` / `scripts/migrate.py`. The deploy script runs migrations before starting the app.
 
 Initial tables:
 - `sessions` — thread_ts → sdk_session_id map
 - `messages` (optional) — conversation history log for debugging
 
 ---
 
-## UC Volume: Same as Cowork
+## UC Volume
 
 **Resource** (`resources/volume.yml`):
 ```yaml
@@ -203,18 +203,15 @@ databricks-claw/
 │   ├── CLAUDE.md                # Personality + instructions (edit this)
 │   ├── core/
 │   │   ├── agent.py             # stream_chat() — claude_agent_sdk.query() loop
+│   │   ├── auth.py              # M2M token generation
 │   │   ├── config.py            # AppConfig — env vars, secrets
-│   │   ├── db.py                # Lakebase engine (copied from Cowork)
+│   │   ├── db.py                # Lakebase engine + token refresh
 │   │   ├── mcp_mapper.py        # UC connection → McpHttpServerConfig
 │   │   ├── memory.py            # MEMORY.md read/write on UC Volume
+│   │   ├── memory_mcp.py        # Memory MCP server
 │   │   ├── sessions.py          # thread_ts → sdk_session_id CRUD
 │   │   ├── slack_client.py      # Slack API calls via UC connection proxy
-│   │   └── slack_events.py      # Event verification + dispatch
-│   ├── alembic/
-│   │   ├── env.py
-│   │   ├── script.py.mako
-│   │   └── versions/
-│   ├── alembic.ini
+│   │   └── slack_poller.py      # Slack polling + dispatch
 │   └── models.py                # SQLModel tables (Session, Message)
 ├── .claude/
 │   └── skills/                  # Skills loaded as plugins into every session
@@ -224,7 +221,9 @@ databricks-claw/
 │   └── volume.yml               # UC Volume for memory
 ├── scripts/
 │   ├── deploy.sh                # Full e2e: build → validate → deploy → migrate → start → app-deploy
-│   └── migrate.py               # Grant SP perms + run Alembic (same as Cowork)
+│   └── migrate.py               # Grant SP perms + run Alembic
+├── tests/                       # Unit tests
+├── alembic.ini
 ├── databricks.yml               # DAB bundle — dev + prod targets + variables
 ├── app.yaml                     # Databricks App runtime config
 └── pyproject.toml
@@ -262,7 +261,7 @@ targets:
 
 ## Deploy Script (e2e)
 
-Same command structure as Cowork's `scripts/deploy.sh`:
+Full e2e deploy via `scripts/deploy.sh`:
 
 ```
 bash scripts/deploy.sh full [--target dev|prod]
@@ -303,14 +302,11 @@ user_api_scopes:
 
 ---
 
-## Open Questions (resolved)
+## Open Questions (all resolved)
 
 - ~~Slack MCP or dedicated connector?~~ → Dedicated connector. Slack is transport, not a Claude tool.
 - ~~OBO or M2M?~~ → M2M with user's Client ID + Secret stored as Databricks Secrets.
 - ~~Memory or stateless?~~ → Both layers: SDK session IDs in Lakebase + MEMORY.md on UC Volume.
 - ~~Streaming to Slack?~~ → Post final response. Can layer in edit-as-streaming later.
-
-## Open Questions (unresolved)
-
-- **Socket Mode vs Events API**: Socket Mode (WebSocket, no public URL) vs Events API (webhook). Socket Mode is simpler for Databricks Apps since no public URL routing needed — but adds a persistent WebSocket connection. Events API requires the app URL to be registered in Slack. Decision needed before implementing `main.py`.
-- **Slack App OAuth scopes**: `chat:write`, `channels:read`, `im:write`, `app_mentions:read` at minimum. Full list depends on what Slack UC connection exposes.
+- ~~Socket Mode vs Events API?~~ → Neither. Uses polling (`slack_poller.py`) — background asyncio task polling `conversations.history`. No public URL or persistent WebSocket needed.
+- ~~Slack App OAuth scopes?~~ → Handled by UC connection configuration. The app never holds a Slack bot token directly.