@defai.digital/automatosx 5.6.10 → 5.6.12

package/CHANGELOG.md

@@ -2,6 +2,36 @@
 
 All notable changes to this project will be documented in this file. See [standard-version](https://github.com/conventional-changelog/standard-version) for commit guidelines.
 
+## [5.6.12](https://github.com/defai-digital/automatosx/compare/v5.6.11...v5.6.12) (2025-10-21)
+
+### Bug Fixes
+
+* **docs:** correct slash command format from `/ax:` to `/ax-` across all files ([#3](https://github.com/defai-digital/automatosx/issues/3)) ([420b17f](https://github.com/defai-digital/automatosx/commit/420b17f))
+  - **Problem**: Command files use dash format (`ax-agent.md`) but documentation incorrectly used colon format (`/ax:agent`)
+  - **Solution**: Corrected all slash commands to use consistent dash format
+  - **Files Changed**: 20 files, 40+ occurrences corrected
+    - **P0**: 7 command definition files (`.claude/commands/*.md`)
+    - **P1**: 1 source code file (`src/cli/commands/init.ts`)
+    - **P2**: 12 documentation files (README, BEST-PRACTICES, examples, etc.)
+  - **Impact**: Users now see correct slash command format (`/ax-agent`, `/ax-status`, `/ax-memory`, etc.) when running `ax init` and reading documentation
+
+### Documentation
+
+* **docs:** update CLAUDE.md for v5.6.11 - 24 agents, workspace conventions ([e8911f7](https://github.com/defai-digital/automatosx/commit/e8911f7))
+  - Updated version from v5.6.9 to v5.6.11
+  - Updated agent count: 19 → 24 agents
+  - Added Phase 2 agent expansion details (Fiona, Ivy)
+  - Consolidated v5.6.9-5.6.11 release notes
+  - Added comprehensive workspace path conventions section
+  - Documented `/tmp` vs `/automatosx/tmp` distinction
+  - Fixed markdown lint warnings
+
+## [5.6.11](https://github.com/defai-digital/automatosx/compare/v5.6.10...v5.6.11) (2025-10-20)
+
+### Version Note
+
+Version 5.6.11 was a version bump to reflect Phase 2 agent expansion completion. All changes are documented in v5.6.10 release notes below.
+
 ## [5.6.10](https://github.com/defai-digital/automatosx/compare/v5.6.9...v5.6.10) (2025-10-20)
 
 

package/README.md

@@ -171,7 +171,7 @@ Use natural language to collaborate with agents directly within your editor. Cla
 "please work with ax agent to refactor this module with best practices"
 ```
 
-For simple, direct tasks, use slash commands: `/ax:agent backend, write a function to validate emails`.
+For simple, direct tasks, use slash commands: `/ax-agent backend, write a function to validate emails`.
 
 ### 2. Terminal/CLI Mode (For Power Users)
 
@@ -208,6 +208,7 @@ ax run quality "Write tests for the API" # Auto-receives design + implementation
 -   **[Core Concepts](docs/guide/core-concepts.md)**
 -   **[Full CLI Command Reference](docs/reference/cli-commands.md)**
 -   **[Agent Directory](examples/AGENTS_INFO.md)**
+-   **[Workspace Path Conventions](docs/workspace-conventions.md)**
 -   **[Troubleshooting Guide](TROUBLESHOOTING.md)**
 
 ---

package/dist/index.js

@@ -5435,8 +5435,8 @@ var initCommand = {
       console.log(chalk27.gray("  \u2022 cto        - Technology strategist"));
       console.log(chalk27.gray("  \u2022 researcher - Research analyst\n"));
       console.log(chalk27.cyan("Claude Code Integration:"));
-      console.log(chalk27.gray("  \u2022 Use /ax command in Claude Code"));
-      console.log(chalk27.gray('  \u2022 Example: /ax assistant "Explain this code"'));
+      console.log(chalk27.gray("  \u2022 Use /ax-agent command in Claude Code"));
+      console.log(chalk27.gray("  \u2022 Example: /ax-agent backend, create a REST API"));
       console.log(chalk27.gray("  \u2022 MCP tools available in .claude/mcp/\n"));
       logger.info("AutomatosX initialized", {
         projectDir,
@@ -11630,6 +11630,23 @@ Note: You can delegate to ANY agent by name, not just those listed above.
 `;
       prompt += `- These are recommendations for organization, not access restrictions
 
+`;
+      prompt += `**File Writing Guidelines:**
+`;
+      prompt += `- ALWAYS use WorkspaceManager for workspace file operations:
+`;
+      prompt += `  \`\`\`typescript
+`;
+      prompt += `  await workspaceManager.writeTmp('report.md', content);
+`;
+      prompt += `  \`\`\`
+`;
+      prompt += `- DO NOT use relative paths like 'tmp/' or './tmp/' for workspace files
+`;
+      prompt += `- DO NOT use direct fs.writeFile() for workspace files (automatosx/tmp/, automatosx/PRD/)
+`;
+      prompt += `- For project files outside workspace (src/, tests/, docs/), use normal file I/O
+
 `;
       prompt += `**When to ask user permission:**
 `;

package/examples/AGENTS_INFO.md

@@ -78,8 +78,8 @@ Instead of directly commanding agents with slash commands, **let Claude Code coo
 ### Express Method: Slash Commands (for simple tasks only)
 
 ```
-⚡ EXPRESS: /ax:agent backend, write an email validation function
-⚡ EXPRESS: /ax:agent quality, review this code snippet
+⚡ EXPRESS: /ax-agent backend, write an email validation function
+⚡ EXPRESS: /ax-agent quality, review this code snippet
 ```
 
 **Use slash commands only when**:

package/examples/claude/commands/ax-agent.md

@@ -1,6 +1,6 @@
 Execute an AutomatosX agent with a specific task.
 
-**IMPORTANT**: When user types `/ax:agent <agent>, <task>`, you MUST:
+**IMPORTANT**: When user types `/ax-agent <agent>, <task>`, you MUST:
 
 1. Split the input on the FIRST comma
 2. Extract agent name (text before comma, trimmed)
@@ -10,7 +10,7 @@ Execute an AutomatosX agent with a specific task.
 **Parsing Rules**:
 
 ```
-Input: /ax:agent backend, explain quantum computing
+Input: /ax-agent backend, explain quantum computing
 ↓
 Agent: "backend"
 Task: "explain quantum computing"
@@ -20,16 +20,16 @@ Execute: automatosx run backend "explain quantum computing"
 
 **Examples**:
 
-User input: `/ax:agent bob, i want you help me write a validation function`
+User input: `/ax-agent bob, i want you help me write a validation function`
 → Execute: `automatosx run bob "i want you help me write a validation function"`
 
-User input: `/ax:agent backend, explain quantum computing to me`
+User input: `/ax-agent backend, explain quantum computing to me`
 → Execute: `automatosx run backend "explain quantum computing to me"`
 
-User input: `/ax:agent backend, create a REST API for user management`
+User input: `/ax-agent backend, create a REST API for user management`
 → Execute: `automatosx run backend "create a REST API for user management"`
 
-User input: `/ax:agent quality, review the changes in src/auth.ts and suggest improvements`
+User input: `/ax-agent quality, review the changes in src/auth.ts and suggest improvements`
 → Execute: `automatosx run quality "review the changes in src/auth.ts and suggest improvements"`
 
 **Available built-in agents**: backend, frontend, devops, data, security, quality, design, writer, product, ceo, cto, researcher

package/examples/claude/commands/ax-clear.md

@@ -1,6 +1,6 @@
 Clear AutomatosX memory.
 
-**IMPORTANT**: When user types `/ax:clear`, you MUST execute:
+**IMPORTANT**: When user types `/ax-clear`, you MUST execute:
 
 ```bash
 automatosx memory clear --confirm
@@ -10,13 +10,13 @@ This will delete all stored memories from the AutomatosX memory database.
 
 **Examples**:
 
-User input: `/ax:clear`
+User input: `/ax-clear`
 → Execute: `automatosx memory clear --confirm`
 
-User input: `/ax:clear --type task`
+User input: `/ax-clear --type task`
 → Execute: `automatosx memory clear --confirm --type task`
 
-User input: `/ax:clear --older-than 30`
+User input: `/ax-clear --older-than 30`
 → Execute: `automatosx memory clear --confirm --older-than 30`
 
 ⚠️ **Warning**: This action cannot be undone. Consider backing up first.

package/examples/claude/commands/ax-init.md

@@ -1,6 +1,6 @@
 Initialize AutomatosX in the current project directory.
 
-**IMPORTANT**: When user types `/ax:init`, you MUST execute:
+**IMPORTANT**: When user types `/ax-init`, you MUST execute:
 
 ```bash
 automatosx init
@@ -17,9 +17,9 @@ This will:
 
 **Examples**:
 
-User input: `/ax:init`
+User input: `/ax-init`
 → Execute: `automatosx init`
 
-User input: `/ax:init --force`
+User input: `/ax-init --force`
 → Execute: `automatosx init --force`
 (Use `--force` to reinitialize if `.automatosx` already exists)

package/examples/claude/commands/ax-list.md

@@ -1,6 +1,6 @@
 List available AutomatosX agents.
 
-**IMPORTANT**: When user types `/ax:list`, you MUST execute:
+**IMPORTANT**: When user types `/ax-list`, you MUST execute:
 
 ```bash
 automatosx list agents
@@ -10,7 +10,7 @@ This will display all available agents with their display names, roles, and desc
 
 **Example**:
 
-User input: `/ax:list`
+User input: `/ax-list`
 → Execute: `automatosx list agents`
 
 Expected output shows all agents including:

package/examples/claude/commands/ax-memory.md

@@ -1,25 +1,25 @@
 Search AutomatosX memory for relevant information.
 
-**IMPORTANT**: When user types `/ax:memory <query>`, you MUST:
+**IMPORTANT**: When user types `/ax-memory <query>`, you MUST:
 
-1. Take everything after `/ax:memory` as the search query
+1. Take everything after `/ax-memory` as the search query
 2. Execute: `automatosx memory search "<query>"`
 
 **Examples**:
 
-User input: `/ax:memory authentication`
+User input: `/ax-memory authentication`
 → Execute: `automatosx memory search "authentication"`
 
-User input: `/ax:memory how to setup database`
+User input: `/ax-memory how to setup database`
 → Execute: `automatosx memory search "how to setup database"`
 
-User input: `/ax:memory API errors and solutions`
+User input: `/ax-memory API errors and solutions`
 → Execute: `automatosx memory search "API errors and solutions"`
 
 **With Options**:
 
-User input: `/ax:memory authentication --limit 5`
+User input: `/ax-memory authentication --limit 5`
 → Execute: `automatosx memory search "authentication" --limit 5`
 
-User input: `/ax:memory database --type task`
+User input: `/ax-memory database --type task`
 → Execute: `automatosx memory search "database" --type task`

package/examples/claude/commands/ax-status.md

@@ -1,6 +1,6 @@
 Display AutomatosX system status and configuration.
 
-**IMPORTANT**: When user types `/ax:status`, you MUST execute:
+**IMPORTANT**: When user types `/ax-status`, you MUST execute:
 
 ```bash
 automatosx status
@@ -16,9 +16,9 @@ This will show:
 
 **Examples**:
 
-User input: `/ax:status`
+User input: `/ax-status`
 → Execute: `automatosx status`
 
-User input: `/ax:status --verbose`
+User input: `/ax-status --verbose`
 → Execute: `automatosx status --verbose`
 (Shows detailed configuration and environment information)

package/examples/claude/commands/ax-update.md

@@ -1,6 +1,6 @@
 Update AutomatosX to the latest version.
 
-**IMPORTANT**: When user types `/ax:update`, you MUST execute:
+**IMPORTANT**: When user types `/ax-update`, you MUST execute:
 
 ```bash
 automatosx update
@@ -15,14 +15,14 @@ This will:
 
 **Examples**:
 
-User input: `/ax:update`
+User input: `/ax-update`
 → Execute: `automatosx update`
 (Interactive update with confirmation)
 
-User input: `/ax:update --check`
+User input: `/ax-update --check`
 → Execute: `automatosx update --check`
 (Only check for updates without installing)
 
-User input: `/ax:update --yes`
+User input: `/ax-update --yes`
 → Execute: `automatosx update --yes`
 (Update without confirmation prompt)

package/examples/pytorch_resnet50_training.py

@@ -0,0 +1,289 @@
+"""
+Mixed precision ResNet-50 training loop for image classification.
+
+Author: Mira — "From architecture to inference - I build models that ship."
+
+This script focuses on the core PyTorch 2.x training primitives:
+- ImageFolder-based DataLoader with realistic augmentations
+- Transfer learning from torchvision's pretrained ResNet-50
+- Mixed precision training via torch.amp.autocast + GradScaler
+- torch.compile() to squeeze extra throughput from the forward pass
+
+Expected directory structure for the dataset:
+data_root/
+  train/
+    class_a/*.jpg
+    class_b/*.jpg
+  val/
+    class_a/*.jpg
+    class_b/*.jpg
+
+Run:
+  python examples/pytorch_resnet50_training.py --data-root /path/to/data
+"""
+
+from __future__ import annotations
+
+import argparse
+import random
+import time
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Tuple
+
+import torch
+from torch import nn
+from torch.amp import GradScaler, autocast
+from torch.optim import Optimizer
+from torch.optim.lr_scheduler import CosineAnnealingLR
+from torch.utils.data import DataLoader
+from torchvision import datasets, transforms
+from torchvision.models import ResNet50_Weights, resnet50
+from tqdm import tqdm
+
+
+@dataclass
+class TrainConfig:
+  data_root: Path
+  batch_size: int = 64
+  num_workers: int = 8
+  epochs: int = 20
+  learning_rate: float = 5e-4
+  weight_decay: float = 0.01
+  seed: int = 17
+  output_dir: Path = Path("artifacts/checkpoints")
+  log_every: int = 25
+  num_classes: int | None = None
+
+
+def set_seed(seed: int) -> None:
+  random.seed(seed)
+  torch.manual_seed(seed)
+  torch.cuda.manual_seed_all(seed)
+  torch.backends.cudnn.deterministic = False
+  torch.backends.cudnn.benchmark = True
+
+
+def build_dataloaders(cfg: TrainConfig) -> Tuple[DataLoader, DataLoader, int]:
+  """Construct train/val DataLoaders and infer class count."""
+  train_dir = cfg.data_root / "train"
+  val_dir = cfg.data_root / "val"
+
+  if not train_dir.is_dir() or not val_dir.is_dir():
+    raise FileNotFoundError(
+        f"Expecting 'train' and 'val' subdirectories under {cfg.data_root}"
+    )
+
+  train_tfms = transforms.Compose(
+      [
+          transforms.RandomResizedCrop(224, scale=(0.6, 1.0)),
+          transforms.RandomHorizontalFlip(),
+          transforms.ColorJitter(0.2, 0.2, 0.2, 0.1),
+          transforms.RandomApply(
+              [transforms.GaussianBlur(kernel_size=3, sigma=(0.1, 2.0))], p=0.3
+          ),
+          transforms.ToTensor(),
+          transforms.Normalize(
+              mean=ResNet50_Weights.IMAGENET1K_V2.meta["mean"],
+              std=ResNet50_Weights.IMAGENET1K_V2.meta["std"],
+          ),
+      ]
+  )
+
+  eval_tfms = transforms.Compose(
+      [
+          transforms.Resize(256),
+          transforms.CenterCrop(224),
+          transforms.ToTensor(),
+          transforms.Normalize(
+              mean=ResNet50_Weights.IMAGENET1K_V2.meta["mean"],
+              std=ResNet50_Weights.IMAGENET1K_V2.meta["std"],
+          ),
+      ]
+  )
+
+  train_dataset = datasets.ImageFolder(train_dir, transform=train_tfms)
+  val_dataset = datasets.ImageFolder(val_dir, transform=eval_tfms)
+  num_classes = len(train_dataset.classes)
+
+  loader_kwargs = dict(
+      batch_size=cfg.batch_size,
+      num_workers=cfg.num_workers,
+      pin_memory=True,
+      persistent_workers=cfg.num_workers > 0,
+  )
+
+  train_loader = DataLoader(train_dataset, shuffle=True, drop_last=True, **loader_kwargs)
+  val_loader = DataLoader(val_dataset, shuffle=False, drop_last=False, **loader_kwargs)
+
+  return train_loader, val_loader, num_classes
+
+
+def build_model(num_classes: int) -> nn.Module:
+  """Load pretrained ResNet-50 and replace the classification head."""
+  base_model = resnet50(weights=ResNet50_Weights.IMAGENET1K_V2)
+  in_features = base_model.fc.in_features
+  base_model.fc = nn.Sequential(
+      nn.Dropout(p=0.2),
+      nn.Linear(in_features, num_classes),
+  )
+  return base_model
+
+
+def accuracy(output: torch.Tensor, target: torch.Tensor) -> torch.Tensor:
+  preds = output.argmax(dim=1)
+  return (preds == target).float().mean()
+
+
+def train_one_epoch(
+    model: nn.Module,
+    loader: DataLoader,
+    optimizer: Optimizer,
+    scaler: GradScaler,
+    device: torch.device,
+    epoch: int,
+    cfg: TrainConfig,
+) -> Tuple[float, float]:
+  model.train()
+  running_loss = 0.0
+  running_acc = 0.0
+
+  for step, (images, labels) in enumerate(tqdm(loader, desc=f"Epoch {epoch} [train]")):
+    images = images.to(device, non_blocking=True)
+    labels = labels.to(device, non_blocking=True)
+
+    optimizer.zero_grad(set_to_none=True)
+
+    with autocast(device_type=device.type, dtype=torch.float16 if device.type == "cuda" else torch.bfloat16):
+      logits = model(images)
+      loss = nn.functional.cross_entropy(logits, labels)
+
+    scaler.scale(loss).backward()
+    scaler.unscale_(optimizer)
+    torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
+    scaler.step(optimizer)
+    scaler.update()
+
+    with torch.no_grad():
+      running_loss += loss.item()
+      running_acc += accuracy(logits, labels).item()
+
+    if step % cfg.log_every == 0:
+      current_lr = optimizer.param_groups[0]["lr"]
+      print(
+          f"Epoch {epoch} | step {step:04d} | lr {current_lr:.2e} | "
+          f"loss {loss.item():.4f}"
+      )
+
+  steps = len(loader)
+  return running_loss / steps, running_acc / steps
+
+
+@torch.no_grad()
+def evaluate(
+    model: nn.Module,
+    loader: DataLoader,
+    device: torch.device,
+) -> Tuple[float, float]:
+  model.eval()
+  total_loss = 0.0
+  total_acc = 0.0
+
+  for images, labels in tqdm(loader, desc="Validation"):
+    images = images.to(device, non_blocking=True)
+    labels = labels.to(device, non_blocking=True)
+
+    logits = model(images)
+    total_loss += nn.functional.cross_entropy(logits, labels).item()
+    total_acc += accuracy(logits, labels).item()
+
+  steps = len(loader)
+  return total_loss / steps, total_acc / steps
+
+
+def save_checkpoint(
+    model: nn.Module,
+    optimizer: Optimizer,
+    epoch: int,
+    cfg: TrainConfig,
+    metric: float,
+) -> None:
+  cfg.output_dir.mkdir(parents=True, exist_ok=True)
+  ckpt = {
+      "epoch": epoch,
+      "state_dict": model.state_dict(),
+      "optimizer_state": optimizer.state_dict(),
+      "val_top1": metric,
+  }
+  torch.save(ckpt, cfg.output_dir / f"resnet50_epoch{epoch:03d}_acc{metric:.3f}.pt")
+
+
+def train_model(cfg: TrainConfig) -> None:
+  set_seed(cfg.seed)
+
+  train_loader, val_loader, inferred_classes = build_dataloaders(cfg)
+  num_classes = cfg.num_classes or inferred_classes
+
+  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+  print(f"Using device: {device}")
+
+  torch.set_float32_matmul_precision("high")
+  model = build_model(num_classes).to(device)
+  model = torch.compile(model)  # PyTorch 2.x graph capture for extra throughput
+
+  optimizer = torch.optim.AdamW(
+      model.parameters(), lr=cfg.learning_rate, weight_decay=cfg.weight_decay
+  )
+  scheduler = CosineAnnealingLR(optimizer, T_max=cfg.epochs)
+  scaler = GradScaler(device.type if device.type == "cuda" else "cpu")
+
+  best_acc = 0.0
+
+  for epoch in range(1, cfg.epochs + 1):
+    epoch_start = time.time()
+
+    train_loss, train_acc = train_one_epoch(
+        model, train_loader, optimizer, scaler, device, epoch, cfg
+    )
+    val_loss, val_acc = evaluate(model, val_loader, device)
+    scheduler.step()
+
+    elapsed = time.time() - epoch_start
+    print(
+        f"Epoch {epoch:02d} finished in {elapsed:.1f}s | "
+        f"train loss {train_loss:.4f}, train acc {train_acc*100:.2f}% | "
+        f"val loss {val_loss:.4f}, val acc {val_acc*100:.2f}%"
+    )
+
+    if val_acc > best_acc:
+      best_acc = val_acc
+      save_checkpoint(model, optimizer, epoch, cfg, metric=val_acc)
+
+
+def parse_args() -> TrainConfig:
+  parser = argparse.ArgumentParser(description="ResNet-50 mixed precision training")
+  parser.add_argument("--data-root", type=Path, required=True, help="Dataset root path")
+  parser.add_argument("--epochs", type=int, default=20)
+  parser.add_argument("--batch-size", type=int, default=64)
+  parser.add_argument("--num-workers", type=int, default=8)
+  parser.add_argument("--lr", type=float, default=5e-4)
+  parser.add_argument("--weight-decay", type=float, default=0.01)
+  parser.add_argument("--seed", type=int, default=17)
+  parser.add_argument("--output-dir", type=Path, default=Path("artifacts/checkpoints"))
+  args = parser.parse_args()
+
+  return TrainConfig(
+      data_root=args.data_root,
+      epochs=args.epochs,
+      batch_size=args.batch_size,
+      num_workers=args.num_workers,
+      learning_rate=args.lr,
+      weight_decay=args.weight_decay,
+      seed=args.seed,
+      output_dir=args.output_dir,
+  )
+
+
+if __name__ == "__main__":
+  config = parse_args()
+  train_model(config)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@defai.digital/automatosx",
-  "version": "5.6.10",
+  "version": "5.6.12",
   "description": "AI Agent Orchestration Platform",
   "type": "module",
   "publishConfig": {