box/docs/AGENT-TOOLING.md

# Agent Tooling & Subagent Delegation Guide

> **Box is the main surface.** All operator work goes through Box (box.muse-dev.online). The web UI, `box` CLI, and agents share the same API endpoints. No UI-only powers.

Welcome, operator. The NetVM environment provides you with the unified `box` command line tool (`/usr/local/bin/box`) for executing tasks, spawning sub-agents, and communicating with peers across the fleet.

---

## 1. Spawning Sub-Agents (`box deploy subagent`)

When you receive a complex task, large audit, or background verification, **prioritize delegating sub-components to an autonomous sub-agent**.

```bash
box deploy subagent --agent <self> --title "<task-name>" "<prompt>"
```

### Parameters:
- `--agent`: Your own identity (`646`, `pip`, `opm`, or `muse`).
- `--title`: Brief descriptive title for the sub-agent session.
- `prompt`: The specific instructions and criteria for the sub-agent.

### Example:
```bash
box deploy subagent --agent 646 --title "exec-api-audit" "Audit the op-based allowlist on exec.muse-dev.online and report back valid ops."
```

---

## 2. Cross-Operator DMs & Relay (`box dm send`)

To coordinate with your peer operators (`opm`, `646`, `pip`, `muse`) or hand off tasks across sidechats:

```bash
box dm send --agent <self> --to <recipient> --target <sidechat> "<message>"
```

### Target Sidechats:
- `646-pip`: Bilateral coordination thread between 646 and Pip.
- `646-opm-coord`: Coordination thread between 646 and OPM.
- `<agent> tasks`: Dedicated task sidechat for an individual agent (e.g. `646 tasks`, `pip tasks`).

### Example:
```bash
box dm send --agent 646 --to pip --target 646-pip "Hey Pip, start-page onboarding review draft is ready. Please confirm when ready to review."
```

---

## 3. Direct Gateway Tooling (`muse` & `box muse`)

You can directly interact with the headless Muse gateway inside your isolated network namespace using either `muse` or `box muse`:

```bash
# Global lookups & fleet status (no account required)
muse status                          # Complete fleet overview & node vitality
muse threads                         # List registered threads and sidechats across fleet
muse unread                          # View unread counts across all agents
muse lookup                          # Unified lookup (summary of fleet, approvals, unread)
muse passkey (or muse key)           # Passkey reference (VM .txt location) & agent approval flow
muse tmux list                       # List shared tmux sessions across fleet

# Using native muse wrapper for per-account actions:
muse -a <account> chat               # Interactive conversational REPL with thread selection
muse -a <account> chat --thread <id> # Direct conversational REPL in specified thread
muse -a <account> status
muse -a <account> threads
muse -a <account> history --thread <thread_uuid> --limit 10
muse -a <account> send --thread <thread_uuid> "<message>"

# If invoked without arguments, it displays available accounts and commands:
muse

# Alternatively via box CLI:
box muse status                      # Cross-fleet status
box muse <self> status               # Agent-specific status
box muse <self> threads              # Active sessions for agent
box muse <self> history --thread <thread_uuid> --limit 10
box muse <self> unread
box muse <self> chat                 # Launch interactive chat REPL
box muse <self> session-start --title "<title>"
box muse <self> send --thread <thread_uuid> "<message>"
box muse tmux list                   # Direct bridge to muse-tmux manager
```

---

## 3.1. Unified & Seamless Lookups (`box lookup` & `box thread`)

For fast inspection of fleet state without hunting across multiple tools:

```bash
# Unified lookup summary (fleet health, pending approvals, key reference)
box lookup
box lookup fleet                     # Node health, CDP status, active pages
box lookup threads                   # List all registered fleet sidechats and mapped UUIDs
box lookup threads <agent>           # List active threads for a specific agent
box lookup unread                    # Unread indicators and active tabs across fleet
box lookup approvals                 # Check if any agent is held on browser approvals
box lookup key (or box passkey)      # Operator passkey & approval protocol

# Seamless thread inspection:
box thread list                      # List all registered fleet sidechats across agents
box thread list <agent>              # List active sessions for an agent
box thread view <agent> <uuid>       # View recent thread messages (supports short UUID prefix)
box thread view <agent> "<alias>"    # View thread by registered alias (e.g. "646 tasks", "heartbeat")
```

---

## 3.2. Operator Passkey & Key Material Architecture

> **Crucial Reality**: Key material and administrative passkeys live in a **single file (`.txt`) on the Google Cloud VM (`34.139.37.135`)**. **NO passkeys or secret stores exist on the dedicated BL (`100.123.153.75`)**.

### Why This Matters:
- Front-door console access (`https://box.muse-dev.online/`) is secured by operator PIN `3128` (or the passkey in the VM text file).
- Operators frequently forget the passkey; it is permanently retrievable via `box passkey` or from the text file on the VM.
- **Agent Rule**: Agents must **NEVER** attempt to grep `bl` or invent imaginary keys. Secrets never reside on the compute node.

### How Agents Get Key Access / Operator Approval:
When an agent or automated task requires elevated privileges, key material, or operator confirmation:
1. **Signal the Request**:
   - In automated scripts: exit with code `2` (the standard `APPROVAL_NEEDED` convention per `INFRA.md`).
   - In sidechats: post `APPROVAL_NEEDED: <details of required key / action>` in the task sidechat (e.g. `646 tasks`, `pip tasks`, `#jobs`, `heartbeat`).
2. **Operator Verification**:
   - The human operator reviews the request in the sidechat or via `box approvals check`.
   - If approved, the operator retrieves the key from the single `.txt` file on the VM (or submits transient OTP via `box cred submit-otp`).
3. **Execution**:
   - The operator authorizes the flow or enters the credential transiently. No raw credentials are saved to `bl` or git.


## 4. Shared & Hybrid Tmux Tooling (`muse tmux`, `box tmux`, & `[TOOL tmux.*]`)

Agents and operators can spawn background sessions and send keystrokes to long-running tasks across three execution tiers:
1. **Central Host**: Runs on the shared agent socket `/tmp/tmux-muse.sock` on `bl`.
2. **Node Network Namespace (NetNS)**: Runs inside isolated node namespaces (`warp-pip`, `warp-dev`, `warp-646`, etc.) using `--node <node>` via `bin/netvm-exec.sh` on `/tmp/tmux-<node>.sock`.
3. **Docker Container**: Runs inside containers using `--container <name>` via `docker exec -i`.

All session stdout/scrollback is automatically piped and persisted to `logs/tmux/<session>[-<target>].log` for auditing and post-mortem analysis.

### Socket Architecture & Strict Isolation
- **Agent Host Socket (`/tmp/tmux-muse.sock`)**: Reserved for agent execution and automated fleet work orders.
- **Operator Desktop Socket (`/tmp/tmux-1000/default`)**: Reserved for user desktop sessions (`main`, `muse`, etc.). Protected in GC routines.
- **Operator LTE Socket (`/tmp/tmux-1000/lte`)**: Reserved for mobile/remote terminal clients.
- **Node NetNS Sockets (`/tmp/tmux-<node>.sock`)**: Isolated inside each node's network namespace.

### Via Native CLI (`box tmux` or `muse-tmux.py`):
```bash
# Host agent session
box tmux new build-worker --command "python3 /srv/worker.py"
box tmux send build-worker "git status"
box tmux capture build-worker --lines 30
box tmux list
box tmux kill build-worker

# Hybrid execution in a specific node netns (e.g. pip, dev, 646)
box tmux --node pip new worker-pip --command "bash"
box tmux --node pip send worker-pip "ip a && hostname"
box tmux --node pip capture worker-pip --lines 20
box tmux --node pip list
box tmux --node pip kill worker-pip

# Hybrid execution inside a Docker container
box tmux --container my-container new worker-c --command "bash"
box tmux --container my-container send worker-c "ps aux"
```

### Via Autonomous Agent Directives:
Agents can emit structured tool calls in sidechats:
```text
[TOOL tmux.new {"session": "build-worker", "command": "python3 /srv/worker.py"}]
[TOOL tmux.send {"session": "build-worker", "keys": "ls -la"}]
[TOOL tmux.capture {"session": "build-worker", "lines": 20}]
[TOOL tmux.list {}]
[TOOL tmux.prune {"ttl": 7200}]
[TOOL tmux.kill {"session": "build-worker"}]
```

---

## 4.1. Agentic Flows in Tmux Panes (`box flow` & `[TOOL flow.*]`)

Chromebox browser contexts prune and store chat history aggressively, making direct in-chat execution of long-running build, test, and shell tasks token-expensive and prone to context loss. 

To overcome this, Chromebox agents offload multi-turn execution to persistent tmux panes on `/tmp/tmux-muse.sock` using the **Flow Engine** (`bin/flow_engine.py`). Raw stdout/stderr streams to disk (`logs/flows/<flow_id>.log`), and agents read back only concise status and incremental output deltas.

### Lifecycle & Primitives:
1. **Start Flow**:
   Spawns pane `flow-<agent>-<id>` and launches command wrapped with an exit code sentinel.
   ```text
   [TOOL flow.start {"flow_id": "audit-tests", "command": "python3 -m unittest discover -s tests"}]
   ```
   *CLI:* `box flow start audit-tests -c "python3 -m unittest discover -s tests"`

2. **Read Incremental Delta & State**:
   Inspects the pane for execution state (`working`, `idle`, `waiting_prompt`, `finished`, `failed`), exit code, and reads newly appended log output since the last read cursor.
   ```text
   [TOOL flow.read {"flow_id": "audit-tests"}]
   ```
   *CLI:* `box flow read audit-tests --lines 40`

3. **Advance or Respond to Prompts**:
   Sends follow-up commands or keystrokes (such as interactive menu selections) without re-running the whole prompt.
   ```text
   [TOOL flow.send {"flow_id": "audit-tests", "command": "git diff"}]
   [TOOL flow.send {"flow_id": "audit-tests", "keys": "1"}]
   ```
   *CLI:* `box flow send audit-tests "git status" --command`

4. **List & Stop**:
   ```text
   [TOOL flow.list {}]
   [TOOL flow.stop {"flow_id": "audit-tests"}]
   ```
   *CLI:* `box flow list` / `box flow stop audit-tests`

---

## 5. Direct Operator Directives & Prompt Envelope Specification

When jobs are dispatched to agents via `bin/job-dispatch.py`, they are wrapped in an actionable, authentic **Operator Directive** generated by `bin/prompt_envelope.py`.

### Directive Structure
- **Header**: Prefixed with `Operator Directive [ref:<ref_id>]:` where `<ref_id>` is a deterministic 8-character identifier derived from the job ID.
- **Work-First Tool Directives**: The directive leads immediately with executable tool calls:
  - Background tmux session setup: `[TOOL tmux.new {"session": "work-<agent>-<ref_id>", "command": "bash"}]`
  - Command transmission: `[TOOL tmux.send {"session": "work-<agent>-<ref_id>", "keys": "echo 'Starting task execution...'"}]`
  - Subagent delegation: `[TOOL swarm.spawn {"count": 2, "task": "..."}]`
  - Scheduled verification: `[TOOL cron.create {"kind": "runonce", ...}]`
- **Naturalized Framing**: Synthetic `[WO:...] WORK ORDER` banners, rigid meta-mandates, and private SSH key signing blocks are completely omitted. This ensures models execute tasks directly without triggering LLM safety tier refusal gates (`Declined — relayed envelope, Tier 2`).
- **Completion Contract**: When task execution finishes, the agent concludes the reply with:
  ```text
  [RESULT <job_id>] OK: <summary of what ran and completed>
  ```
  The harvester (`bin/response-harvester.py`) detects this line and records the job outcome.

---

## 6. Best Practices & Invariants

1. **Sidechat-First Policy**: All inter-agent coordination, subagent tasks, and heartbeats must stay in **sidechats**. Do not send automated routine messages to `main` chat (see [CHAT_POLICY.md](file:///home/super/Projects/NetVM/CHAT_POLICY.md)).
2. **Sub-Agent Prioritization**: Break down complex diagnostic or verification jobs by delegating sub-tasks to dedicated subagent threads.
3. **Execution Reality**: Work is only real if tool calls ran. Never provide purely verbal confirmation for tasks requiring system inspection or execution.
4. **Attribution & Result Tagging**: For scheduled jobs and work orders, always conclude your response with `[RESULT <job_id>] <summary>`.

---

## 7. Internal Documentation & Agent Lookups (`box docs` & `docs_internal/`)

Agents have access to a structured internal `.md` and `.json` database in `docs_internal/` for looking up agent sentence structures, regex passing, and assistive surfaces for `box.muse-dev.online`:

```bash
# Query surfaces, DOM selectors, and REST endpoints for box.muse-dev.online
box docs surfaces jobs
box docs surfaces dms

# Inspect agent sentence structures and conversational contracts
box docs sentence work_order
box docs sentence result

# Test strings or evaluate against canonical regex patterns
box docs regex result --test "[RESULT 7fce46e0] OK 14 endpoints verified"
box docs parse "[WO:7fce46e0] [from super] Audit exec — Check stats"

# Full-text search across documentation database
box docs search "work order"
```