pzsh 0.3.5

Performance-first shell framework with sub-10ms startup
Documentation
# Completion

pzsh provides intelligent auto-completion with optional ML-based predictions.

## Configuration

Enable completion in `~/.pzshrc`:

```toml
[completion]
enabled = true
# Optional: ML model for predictions
# model_path = "~/.pzsh/models/aprender-shell.onnx"
```

## Completion Sources

### Alias Completion

Completes configured aliases with expansion preview:

```
$ gs<TAB>
gs → git status
gp → git push
```

### Path Completion

Completes file and directory paths:

```
$ cd /usr/l<TAB>
/usr/lib
/usr/local
```

### Environment Variables

Completes environment variables:

```
$ echo $HO<TAB>
$HOME
$HOSTNAME
```

### History Completion

Completes from command history:

```
$ git com<TAB>
git commit -m 'fix typo'
git commit --amend
```

### Command Completion

Built-in completions for common commands:

- **git**: Subcommands, branches, remotes
- **docker**: Containers, images, commands
- **cargo**: Subcommands, packages

## Zsh Integration

pzsh generates native zsh completion functions:

```zsh
# Generated by pzsh compile
autoload -Uz compinit
compinit -C -d "${ZDOTDIR:-$HOME}/.zcompdump"

# Completion styling
zstyle ':completion:*' menu select
zstyle ':completion:*' matcher-list 'm:{a-zA-Z}={A-Za-z}'
```

## ML-Based Predictions

When configured with an aprender-shell model:

```toml
[completion]
model_path = "~/.pzsh/models/aprender-shell.onnx"
```

The completion engine will:
1. Use traditional completions as fallback
2. Score suggestions using the ML model
3. Rank by predicted likelihood
4. Stay within 50ms budget

## Performance

- Completion lookup: O(1) via AHashMap
- Path scanning: Bounded to 100 entries
- ML inference: 50ms budget with fallback
- No blocking operations in critical path