### Development Installation Command

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/packages/ara-skills/README.md

Commands to set up the development environment for ara-skills, including installing dependencies and performing a local installation.

```bash
cd packages/ara-skills
npm install
node bin/cli.js install --skill compiler --local --force
```

--------------------------------

### Install ARA Skills Locally

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/README.md

Install all ARA skills into the current project directory instead of the user's home directory.

```bash
npx @orchestra-research/ara-skills install --all --local
```

--------------------------------

### ARA Skills Install Flags

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Options for the `ara-skills install` command to specify which skills and agents to install, and installation scope.

```bash
--all                    Install all three skills (default if no --skill given)
```

```bash
--skill <id>             Specify a skill (repeatable); ids: compiler, research-manager, rigor-reviewer
```

```bash
--agent <id>             Specify an agent (repeatable); ids: claude-code, cursor, gemini-cli,
                         opencode, codex, hermes, generic
```

```bash
--local                  Install into ./<agent-dir>/skills (project scope) instead of $HOME
```

```bash
--force                  Overwrite existing installations
```

```bash
--quiet                  Suppress per-skill log output
```

--------------------------------

### Clone Repository and Install CLI Dependencies

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Steps to clone the project repository and install the necessary npm dependencies for the `ara-skills` CLI package.

```bash
# Clone the repository
git clone https://github.com/Orchestra-Research/Agent-Native-Research-Artifact-Init.git
cd Agent-Native-Research-Artifact-Init

# Install dependencies for the CLI package
cd packages/ara-skills
npm install
```

--------------------------------

### Debug ARA Skills Installation

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Example of enabling debug mode for `ara-skills install` by setting the ARA_SKILLS_DEBUG environment variable.

```bash
ARA_SKILLS_DEBUG=1 npx @orchestra-research/ara-skills install --all
```

--------------------------------

### Install All ARA Skills Globally

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/packages/ara-skills/README.md

Installs all available ARA skills to every detected agent on your system.

```bash
npx @orchestra-research/ara-skills install --all
```

--------------------------------

### Install ARA Skills Globally

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/README.md

Install all ARA skills globally using npx. This command auto-detects compatible agents and prompts for installation scope.

```bash
npx @orchestra-research/ara-skills
```

--------------------------------

### Install ARA Skills with npm CLI

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Use the npx command to install ARA skills. The CLI auto-detects compatible agents and installs skills accordingly. Options include interactive install, installing all skills globally or locally, and specifying skills or agents.

```bash
npx @orchestra-research/ara-skills
```

```bash
npx @orchestra-research/ara-skills install --all
```

```bash
npx @orchestra-research/ara-skills install --skill compiler --agent claude-code
```

```bash
npx @orchestra-research/ara-skills install --all --local
```

```bash
npx @orchestra-research/ara-skills install \
  --skill compiler \
  --skill rigor-reviewer \
  --agent claude-code \
  --agent cursor \
  --force
```

```bash
npx @orchestra-research/ara-skills list
```

```bash
npx @orchestra-research/ara-skills update
```

```bash
npx @orchestra-research/ara-skills uninstall --skill rigor-reviewer --agent cursor
```

```bash
npx @orchestra-research/ara-skills skills
```

```bash
npx @orchestra-research/ara-skills agents
```

--------------------------------

### SKILL.md Frontmatter Example

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/CONTRIBUTING.md

YAML frontmatter for SKILL.md, including name, description, argument hint, allowed tools, and metadata. Ensure the description is concise and in the third person.

```yaml
---
name: my-skill
description: >
  What it does AND when to use it. Include trigger keywords.
  Max 1024 chars. Third person.
argument-hint: "[optional args description]"
allowed-tools: Read, Write, Edit, Glob, Grep
metadata:
  author: your-name
  version: "1.0.0"
  tags: [keyword1, keyword2]
---

```

--------------------------------

### Install ARA Skills Programmatically

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Use the installer module to manage skills for agents. Import functions like `install`, `installMany`, `update`, `uninstall`, and `listInstalled` from '@orchestra-research/ara-skills/src/installer.js'.

```javascript
import { install, installMany, uninstall, update, listInstalled } from '@orchestra-research/ara-skills/src/installer.js';

// Install all skills to claude-code (global)
const result = install({ agentId: 'claude-code', skillIds: [], local: false, force: false });
// result: {
//   agent: 'claude-code',
//   targetDir: '/Users/alice/.claude/skills',
//   results: [
//     { skill: 'compiler',          status: 'installed', dest: '…/compiler' },
//     { skill: 'research-manager',  status: 'installed', dest: '…/research-manager' },
//     { skill: 'rigor-reviewer',    status: 'installed', dest: '…/rigor-reviewer' }
//   ]
// }

// Install a single skill to a local project (into ./.cursor/skills/)
install({ agentId: 'cursor', skillIds: ['compiler'], local: true, force: true, quiet: true });

// Install to multiple agents in one call
installMany({
  agentIds: ['claude-code', 'cursor', 'gemini-cli'],
  skillIds: ['compiler', 'rigor-reviewer'],
  local: false,
  force: false,
});

// Update all tracked skills (re-install with force) for an agent
update({ agentId: 'claude-code', local: false });

// Uninstall all skills from claude-code
uninstall({ agentId: 'claude-code', skillIds: [], local: false });

// Uninstall one skill
uninstall({ agentId: 'claude-code', skillIds: ['research-manager'] });

// List what is installed, across all agents and scopes
const rows = listInstalled();
// rows: [
//   { agent: 'claude-code', scope: 'global', dir: '…', skills: ['compiler', 'rigor-reviewer'], tracked: true }
// ]
```

--------------------------------

### Generate Full Artifact with Compiler

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/examples/README.md

To generate a full artifact, install the compiler skill and run this command, providing the path to your PDF.

```bash
/compiler path/to/paper.pdf
```

--------------------------------

### Exploration Tree Node Example

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

An example of an exploration tree node being written directly, indicating that decisions are directly routed rather than staged.

```yaml
# Example: exploration tree node written directly (decisions are direct-routed)
```

--------------------------------

### Run ARA Skills CLI in Development Mode

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Instructions for running the `ara-skills` CLI locally during development, pointing to monorepo skills and installing to a project-local directory.

```bash
# Run the CLI in dev mode (reads skills from ../../skills/ in the monorepo)
node bin/cli.js install --skill compiler --local --force
# Installs from: skills/compiler/ (monorepo root)
# Into: .claude/skills/compiler (project-local)
```

--------------------------------

### Installer Module Functions

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Functions for managing ARA skills programmatically within Node.js tooling.

```APIDOC
## Installer Module API

### Description

The installer module exports functions for embedding skill management in Node.js tooling.

### Functions

- **`install(options)`**: Installs skills for a specified agent.
  - `options` (object): Configuration for installation.
    - `agentId` (string): The ID of the agent to install skills for.
    - `skillIds` (array): An array of skill IDs to install. An empty array installs all skills.
    - `local` (boolean): If true, installs to a local project directory (`./.cursor/skills/`).
    - `force` (boolean): If true, forces re-installation of existing skills.
    - `quiet` (boolean, optional): If true, suppresses output.
  - Returns: An object detailing the installation results.

- **`installMany(options)`**: Installs skills to multiple agents simultaneously.
  - `options` (object): Configuration for bulk installation.
    - `agentIds` (array): An array of agent IDs.
    - `skillIds` (array): An array of skill IDs to install.
    - `local` (boolean): If true, installs to local project directories.
    - `force` (boolean): If true, forces re-installation.
  - Returns: An object detailing the installation results.

- **`update(options)`**: Updates all tracked skills for an agent (re-installs with force).
  - `options` (object): Configuration for updating skills.
    - `agentId` (string): The ID of the agent.
    - `local` (boolean): If true, updates local installations.
  - Returns: An object detailing the update results.

- **`uninstall(options)`**: Uninstalls skills from an agent.
  - `options` (object): Configuration for uninstallation.
    - `agentId` (string): The ID of the agent.
    - `skillIds` (array): An array of skill IDs to uninstall. An empty array uninstalls all skills.
    - `local` (boolean): If true, uninstalls from local project directories.
  - Returns: An object detailing the uninstallation results.

- **`listInstalled()`**: Lists all installed skills across all agents and scopes.
  - Returns: An array of objects, each describing an installed agent scope and its skills.

### Example Usage

```javascript
import { install, installMany, uninstall, update, listInstalled } from '@orchestra-research/ara-skills/src/installer.js';

// Install all skills to claude-code (global)
const result = install({ agentId: 'claude-code', skillIds: [], local: false, force: false });

// Install a single skill to a local project
install({ agentId: 'cursor', skillIds: ['compiler'], local: true, force: true, quiet: true });

// Install to multiple agents
installMany({
  agentIds: ['claude-code', 'cursor', 'gemini-cli'],
  skillIds: ['compiler', 'rigor-reviewer'],
  local: false,
  force: false,
});

// Update all tracked skills for an agent
update({ agentId: 'claude-code', local: false });

// Uninstall all skills from claude-code
uninstall({ agentId: 'claude-code', skillIds: [], local: false });

// Uninstall one skill
uninstall({ agentId: 'claude-code', skillIds: ['research-manager'] });

// List installed skills
const rows = listInstalled();
```
```

--------------------------------

### Install Specific ARA Skill and Agent

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/README.md

Install a particular skill (e.g., compiler) for a specific agent (e.g., claude-code).

```bash
npx @orchestra-research/ara-skills install --skill compiler --agent claude-code
```

--------------------------------

### ARA Evidence Naming Convention Example

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/skills/compiler/references/ara-schema.md

Examples illustrating the naming conventions for evidence files, distinguishing between raw source evidence and derived subset evidence. Raw evidence must preserve original identifiers, while derived evidence must indicate its origin.

```text
evidence/tables/table3_imagenet_validation.md
evidence/tables/derived_from_table3_residual_depth_slice.md
```

--------------------------------

### Crystallized Heuristic Example

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

An example of a crystallized heuristic, promoted from a staged observation after user affirmation. It includes rationale, provenance, and references to staging and code.

```yaml
# Example: crystallized heuristic (after user says "yes, let's document that")
# ara/logic/solution/heuristics.md
## H04: Weight decay sensitivity for deep ResNets
- **Rationale**: wd=1e-3 causes gradient explosion at depth ≥50; wd=1e-4 is the stable boundary.
- **Provenance**: user-revised
- **Crystallized via**: verbal-affirmation
- **Sensitivity**: high
- **Code ref**: [src/execution/training_recipe.py]
- **From staging**: O07
```

--------------------------------

### Manual Installation of ARA Skills for Claude Code

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Skills can be manually copied into the agent's skills directory if the npm CLI is not used. This allows for project-level or user-level installation of all or specific skills.

```bash
cp -r skills/* .claude/skills/
```

```bash
cp -r skills/* ~/.claude/skills/
```

```bash
cp -r skills/compiler ~/.claude/skills/compiler
```

```bash
# Verify the structure Claude Code expects:
# ~/.claude/skills/
#   compiler/
#     SKILL.md
#     references/
#       ara-schema.md
#       exploration-tree-spec.md
#       validation-checklist.md
#   research-manager/
#     SKILL.md
#     references/
#       event-taxonomy.md
#   rigor-reviewer/
#     SKILL.md
#     references/
#       review-dimensions.md
```

--------------------------------

### Staged Observation Example

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

An example of a staged observation created during an agent's turn, which is not yet crystallized. It includes details like provenance, content, and context.

```yaml
# Example: staged observation created during a turn
# ara/staging/observations.yaml
observations:
  - id: O07
    timestamp: "2025-06-10T14:32"
    provenance: ai-suggested
    content: "Weight decay 1e-4 is critical; 1e-3 causes divergence after epoch 5."
    context: "Training run log showed NaN loss with wd=1e-3 at step 2000."
    potential_type: heuristic
    bound_to: [N03]
    promoted: false
    promoted_to: null
    crystallized_via: null
    stale: false
```

--------------------------------

### ARA Skills CLI Commands

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/packages/ara-skills/README.md

Overview of available commands for managing ARA skills, including installation, updates, and listing.

```bash
ara-skills                             # interactive
ara-skills install  [--all] [--skill <id>] [--agent <id>] [--local] [--force]
ara-skills update   [--agent <id>] [--local]
ara-skills uninstall [--skill <id>] [--agent <id>] [--local]
ara-skills list                        # what is installed, where
ara-skills skills                      # what's available to install
ara-skills agents                      # which agents are supported / detected
```

--------------------------------

### Manual Install ARA Skills for Claude Code

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/README.md

Manually copy ARA skills into the Claude Code configuration directory for project-level or user-level access.

```bash
cp -r skills/* .claude/skills/
```

```bash
cp -r skills/* ~/.claude/skills/
```

```bash
cp -r skills/compiler ~/.claude/skills/compiler
```

--------------------------------

### Bundle Skills for npm Publish

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Commands related to the `prepack` and `postpack` scripts, which bundle skills into the npm tarball for publishing.

```bash
# The prepack / postpack scripts bundle skills for npm publish:
#   prepack: copies skills/ into packages/ara-skills/skills/ (makes tarball self-contained)
#   postpack: removes the copy
npm pack   # produces @orchestra-research-ara-skills-0.2.0.tgz with bundled skills
```

--------------------------------

### Initialize Research Artifact Directory Structure

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/skills/research-manager/SKILL.md

Command to create the necessary directory structure for a new research artifact when the 'ara/' directory does not yet exist. This should be run on the first turn with research-significant activity.

```bash
mkdir -p ara/{logic/solution,src/{configs,kernel},trace/sessions,evidence/{tables,figures},staging}
```

--------------------------------

### Skill Directory Structure

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/CONTRIBUTING.md

Standard directory layout for a new skill. SKILL.md is required, while references/ and templates/ are optional.

```bash
skills/my-skill/
├── SKILL.md              # Required — main instructions
├── references/           # Optional — detailed docs loaded on demand
└── templates/            # Optional — starter templates
```

--------------------------------

### Rigor Reviewer Skill Invocation

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Example of invoking the rigor-reviewer skill on an ARA artifact directory for semantic review.

```text
# Invoked on an ARA artifact directory:
/rigor-reviewer examples/resnet-ara-example/

# Produces: examples/resnet-ara-example/level2_report.json
```

--------------------------------

### Run ARA Compiler with Input

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/README.md

Use the compiler skill to convert various research inputs like PDFs or GitHub repositories into ARA artifacts. Specify an output directory if needed.

```bash
/compiler path/to/paper.pdf
```

```bash
/compiler https://github.com/org/repo
```

```bash
/compiler path/to/paper.pdf path/to/code/ --output ./my-artifact/
```

--------------------------------

### ARA Artifact Output Structure

Source: https://context7.com/orchestra-research/agent-native-research-artifact/llms.txt

Example structure of the output produced by the /compiler skill, detailing the organization of research components like paper content, logic, source code, and trace information.

```text
ara-output/
  PAPER.md                            # YAML frontmatter + Layer Index (~200 tokens)
  logic/
    problem.md                        # O1/O2 observations → G1/G2 gaps → key insight → assumptions
    claims.md                         # C01–C{N} falsifiable claims with proof refs to E-IDs
    concepts.md                       # ≥5 formal definitions with notation + boundary conditions
    experiments.md                    # E01–E{N} declarative plans (directional outcomes only)
    solution/
      architecture.md                 # Component graph with inputs/outputs
      algorithm.md                    # Math formulation + pseudocode + complexity
      constraints.md                  # Boundary conditions and known limitations
      heuristics.md                   # H01–H{N} implementation tricks with sensitivity + code refs
    related_work.md                   # RW01–RW{N} typed dependency graph
  src/
    configs/
      training.md                     # Hyperparameters with exact values, rationale, sensitivity
      model.md                        # Architecture configs
    execution/
      {module}.py                     # Typed Python stubs for the novel contribution
    environment.md                    # Python version, framework, hardware, seeds
  trace/
    exploration_tree.yaml             # Research DAG ≥8 nodes with dead_end and decision types
  evidence/
    README.md                         # Index mapping every evidence file to claim IDs
    tables/
      table2_imagenet_plain_vs_residual.md   # Raw source table (exact cell values)
      derived_from_table3_shortcut_options.md # Derived subset (labeled as such)
    figures/
      figure1_cifar_plain_curves.md   # Extracted data points with axes
```

--------------------------------

### ResNet Training Recipe in Python

Source: https://github.com/orchestra-research/agent-native-research-artifact/blob/main/examples/resnet-ara-example/PAPER.md

Implements a standard training recipe for ImageNet ResNets using SGD with a step learning rate schedule and Batch Normalization after convolution. This is suitable for training deep ResNet models on large datasets.

```python
class TrainingRecipe(nn.Module):
    def __init__(self, block, num_blocks, num_classes=10, num_channels=64):
        super(TrainingRecipe, self).__init__()
        self.in_planes = num_channels
        self.conv1 = nn.Conv2d(3, num_channels, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(num_channels)
        self.layer1 = self._make_layer(block, num_channels, num_blocks[0], stride=1)
        self.layer2 = self._make_layer(block, num_channels*2, num_blocks[1], stride=2)
        self.layer3 = self._make_layer(block, num_channels*4, num_blocks[2], stride=2)
        self.layer4 = self._make_layer(block, num_channels*8, num_blocks[3], stride=2)
        self.linear = nn.Linear(num_channels*8*block.expansion, num_classes)

    def _make_layer(self, block, planes, num_blocks, stride):
        strides = [stride] + [1]*(num_blocks-1)
        for stride in strides:
            layers = []
            if stride != 1 or self.in_planes != planes * block.expansion:
                layers.append(block(self.in_planes, planes, stride))
            else:
                layers.append(block(self.in_planes, planes, stride))
            self.in_planes = planes * block.expansion
        return nn.Sequential(*layers)

    def forward(self, x):
        out = F.relu(self.bn1(self.conv1(x)))
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        out = self.layer4(out)
        out = F.avg_pool2d(out, 10)
        out = out.view(out.size(0), -1)
        out = self.linear(out)
        return out

def ResNet18(num_classes=10, num_channels=64):
    return TrainingRecipe(BasicBlock, [2,2,2,2], num_classes=num_classes, num_channels=num_channels)

def ResNet34(num_classes=10, num_channels=64):
    return TrainingRecipe(BasicBlock, [3,4,6,3], num_classes=num_classes, num_channels=num_channels)

def ResNet50(num_classes=10, num_channels=64):
    return TrainingRecipe(Bottleneck, [3,4,6,3], num_classes=num_classes, num_channels=num_channels)

def ResNet101(num_classes=10, num_channels=64):
    return TrainingRecipe(Bottleneck, [3,4,23,3], num_classes=num_classes, num_channels=num_channels)

def ResNet152(num_classes=10, num_channels=64):
    return TrainingRecipe(Bottleneck, [3,8,36,3], num_classes=num_classes, num_channels=num_channels)

def آموزش_شبکه_عمیق(model, device, train_loader, optimizer, criterion, epochs=100):
    model.train()
    for epoch in range(epoch, epochs + 1):
        running_loss = 0.0
        for i, data in enumerate(train_loader, 0):
            inputs, labels = data
            inputs, labels = inputs.to(device), labels.to(device)
            optimizer.zero_grad()
            outputs = model(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()
            running_loss += loss.item()
            if i % 2000 == 1999:
                print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 2000:.3f}')
                running_loss = 0.0
        print(f'Finished Epoch {epoch + 1}')
    print('Finished Training')

def ارزیابی_مدل(model, device, test_loader, criterion):
    model.eval()
    correct = 0
    total = 0
    with torch.no_grad():
        for data in test_loader:
            images, labels = data
            images, labels = images.to(device), labels.to(device)
            outputs = model(images)
            loss = criterion(outputs, labels)
            _, predicted = torch.max(outputs.data, 1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    print(f'Accuracy of the network on the 10000 test images: {100 * correct // total} %')
    print(f'Test Loss: {loss.item():.3f}')
```