Hooks Overview

---
name: hooks-overview
description: Lifecycle hooks that log edits, outcomes and trajectories to local pattern files and route tasks to agents.
type: flow
---

# Hooks System Overview

Lifecycle hooks that connect Claude Code tool events to Monomind's outcome logging, agent routing, and session persistence. Hooks log edits, outcomes and trajectories to local JSON pattern files; no model is trained.

## How It Works

Claude Code fires hook events (PreToolUse, PostToolUse, etc.) which trigger `npx monomind hooks <subcommand>` commands. The hooks system:

1. **Routes** tasks to agents through the keyword picker (pick stats act as a bounded ranking prior)
2. **Records** outcomes (edits, commands, tasks) in local JSON pattern files
3. **Persists** session state across conversations
4. **Scans** the repository (file types, import lines) into the memory store and local pattern log (`hooks pretrain`)

## CLI Subcommands

All hooks are invoked as `npx monomind hooks <subcommand>`:

### Lifecycle Hooks
| Subcommand | Purpose |
|---|---|
| `pre-edit` | Context + agent suggestions before file edit |
| `post-edit` | Record edit outcome |
| `pre-command` | Risk assessment before running a command |
| `post-command` | Record command outcome |
| `pre-task` | Register task start, get agent suggestions + model routing |
| `post-task` | Record task completion |
| `session-end` | End session and persist state |
| `session-restore` | Restore a previous session |

### Intelligence & Routing
| Subcommand | Purpose |
|---|---|
| `route` | Route task to optimal agent |
| `explain` | Explain routing decision |
| `pretrain` | Scan the repository (file types, import lines) into the memory store and local pattern log (no model is trained) |
| `metrics` | View recorded routing/outcome metrics |
| `model-route` | Route to optimal model (haiku/sonnet/opus) |
| `model-outcome` | Record model routing result |
| `model-stats` | View model routing statistics |

### Coverage Tools
| Subcommand | Purpose |
|---|---|
| `coverage-route` | Route based on test coverage gaps |
| `coverage-suggest` | Suggest coverage improvements |
| `coverage-gaps` | List all coverage gaps with priorities |

### Workers & Utilities
| Subcommand | Purpose |
|---|---|
| `worker` | Background worker management (<!-- doc-count:workers -->9<!-- /doc-count:workers --> workers; `worker list`, `worker run <name>`) |
| `intelligence` | JS pattern/trajectory store (train, patterns, predict, optimize, export, import) |
| `notify` | Send a notification |
| `statusline` | Generate dynamic statusline display |
| `list` | List all registered hooks |
| `transfer` | Transfer patterns from another local project |

## Claude Code Integration

Configure in `.claude/settings.json`:

```json
{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "^(Write|Edit|MultiEdit)$",
        "hooks": [{
          "type": "command",
          "command": "npx monomind hooks pre-edit --file '${tool.params.file_path}'"
        }]
      },
      {
        "matcher": "^Bash$",
        "hooks": [{
          "type": "command",
          "command": "npx monomind hooks pre-command --command '${tool.params.command}'"
        }]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "^(Write|Edit|MultiEdit)$",
        "hooks": [{
          "type": "command",
          "command": "npx monomind hooks post-edit --file '${tool.params.file_path}' --success true"
        }]
      }
    ]
  }
}
```

## 4-Step Intelligence Pipeline (pretrain)

Running `npx monomind hooks pretrain` executes:
1. **RETRIEVE** — Top-k memory injection with MMR diversity
2. **JUDGE** — LLM-as-judge trajectory evaluation
3. **DISTILL** — Extract strategy memories from trajectories
4. **CONSOLIDATE** — Dedup, detect contradictions, prune old patterns

Optionally adds:
5. **EMBED** — Index documents with ONNX model
6. **HYPERBOLIC** — Poincaré ball projection for hierarchy preservation