> ## Documentation Index
> Fetch the complete documentation index at: https://www.agentworldprotocol.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Envelopes

> Quantitative safety limits, enforced server-side.

The world's `safety_policy` declares envelopes per embodiment:

```json awp:envelope theme={null}
{
  "embodiment": "arm_01",
  "spatial": { "frame": "base", "aabb_m": [[-0.8,-0.8,0.0],[0.8,0.8,1.2]] },
  "max_velocity_mps": 1.0,
  "max_force_n": 40,
  "max_action_rate_hz": 10,
  "enforcement": "both",
  "on_violation": "clamp"
}
```

* Enforcement is **world-side and unconditional**: no agent input may cause envelope violation. `[AWP-ENV-001]`
* `on_violation` is `clamp` (execute at the limit, report `clamped: true` in status) or `reject` (`AWP_ENVELOPE_EXCEEDED`). Declared per envelope. `[AWP-ENV-002]`
* Violations of physical envelopes during execution (external disturbance) MUST fail the action (reason `envelope`) and emit `world.event: envelope_violation`. `[AWP-ENV-003]`
* Rate limiting applies to `action.submit` admission; excess returns `AWP_ENVELOPE_EXCEEDED` with `retryable: true` and `retry_after_ms`. `[AWP-ENV-004]`
* Each envelope declares `enforcement`: `command_check` (requested commands are clamped or rejected before actuation; the default), `measured` (executed motion is monitored and violations fail the action per AWP-ENV-003), or `both`. A world MUST NOT declare a limit it cannot enforce at least by `command_check`, and MUST NOT declare `measured` or `both` for a quantity it cannot observe. The [robotics profile](/spec/profiles/robotics) fixes which mode each limit requires. `[AWP-ENV-005]`

On [command channels](/spec/loop/command-channels), enforcement applies per inbound frame (AWP-CMD-006).