# {{name}}

{{description}}

## What this harness does

Coordinates on-call incident response: alert ingestion, runbook lookup, escalation logic, and postmortem authoring. 4 agents work in a hierarchical pipeline; the kernel routes between them.

## Agents

| Agent | Role | Typical model tier |
|---|---|---|
| `responder` | Triages incoming alerts, identifies playbook | Haiku |
| `runbook-runner` | Executes known runbooks step by step | Haiku |
| `escalator` | Decides when to page humans + who | Sonnet |
| `postmortem` | Drafts incident postmortems from session transcript | Sonnet |

## Behavioral rules

- ALWAYS run `runbook-runner` before `escalator` — escalation requires runbook context
- NEVER auto-execute destructive commands; runbook-runner deliberates, doesn't `rm`
- `escalator` may invoke a human via the `notify` MCP tool — that's the only side-effect agent

## Slash commands

| Command | What it does |
|---|---|
| `/triage <alert>` | Run responder + runbook-runner |
| `/escalate <reason>` | Force escalation path |
| `/postmortem` | Generate postmortem from current session |

## Architecture

Built on `@metaharness/kernel` (Rust + WASM). Multi-host: ships with the **{{host}}** adapter; other hosts available via `harness add-host <name>`.