OpenAdapt Bootstrap

Self-hosting infrastructure for OpenAdapt recursive development

Concept

OpenAdapt Bootstrap enables OpenAdapt to build OpenAdapt - a recursive self-improvement system where recorded development workflows can be replayed autonomously via Claude Code integration.

This creates infrastructure for:

• Automated demo generation
• Self-documenting workflows
• Reproducible development tasks
• Mobile-first development (user directs from mobile, desktop executes autonomously)

Architecture

openadapt-bootstrap/
├── recordings/           # OpenAdapt recordings of dev workflows
│   ├── demo_generation/
│   ├── screenshot_capture/
│   ├── test_execution/
│   └── pr_creation/
├── workflows/            # High-level workflow definitions
│   ├── __init__.py
│   ├── base.py          # WorkflowRecorder, WorkflowExecutor base classes
│   ├── demo_generation.py
│   ├── screenshot_workflow.py
│   └── test_workflow.py
├── playback/            # Replay automation scripts
│   ├── __init__.py
│   └── executor.py      # Claude Code integration
├── examples/            # Example workflow usage
│   └── generate_benchmark_screenshots.py
├── docs/
│   └── ARCHITECTURE.md
└── README.md

Key Components

1. Workflow Recorder

Captures development tasks as OpenAdapt recordings:

from openadapt_bootstrap import WorkflowRecorder

# Record a workflow once (manual demonstration)
with WorkflowRecorder(
    name="generate_pr_screenshots",
    description="Generate screenshots for benchmark viewer PR"
) as recorder:
    # Perform the task manually...
    # - Open browser to benchmark viewer
    # - Navigate to different states
    # - Capture screenshots
    # - Save to screenshots/ directory
    pass

print(f"Recorded workflow: {recorder.recording_path}")

2. Workflow Executor

Replays workflows autonomously via Claude Code:

from openadapt_bootstrap import WorkflowExecutor

# Replay workflow without user intervention
executor = WorkflowExecutor(
    recording_path="recordings/demo_generation/pr_screenshots.db",
    claude_code_enabled=True
)

result = executor.execute()
print(f"Screenshots saved to: {result.output_dir}")

3. Claude Code Integration

Execute workflows in the background while user is on mobile:

from openadapt_bootstrap import ClaudeCodeWorkflow

# Trigger from GitHub issue comment (user on mobile)
workflow = ClaudeCodeWorkflow.from_github_issue(
    repo="OpenAdaptAI/openadapt-evals",
    issue_number=42
)

# Execute autonomously
result = workflow.execute()

# Post results back to GitHub
workflow.post_result_to_issue(result)

Use Cases

1. Auto-Generate PR Screenshots

Problem: Creating screenshots for PRs is manual and time-consuming

Solution: Record the screenshot workflow once, replay for each PR

from openadapt_bootstrap.workflows import ScreenshotWorkflow

workflow = ScreenshotWorkflow(
    target_url="http://localhost:8080/viewer.html",
    viewports=["desktop", "tablet", "mobile"],
    states=["overview", "task_detail", "log_expanded"]
)

screenshots = workflow.execute()
workflow.commit_and_push(screenshots, branch="pr-screenshots")

2. Automated Demo Generation

Problem: Demos need to be regenerated when UI changes

Solution: Record demo workflow, replay on demand

from openadapt_bootstrap.workflows import DemoGenerationWorkflow

workflow = DemoGenerationWorkflow(
    demo_script="scripts/notepad_demo.py",
    output_format="gif",
    duration_seconds=15
)

demo_path = workflow.execute()

3. Test Suite Execution with Visual Verification

Problem: Tests pass but UI might be broken

Solution: Record test execution with visual checkpoints

from openadapt_bootstrap.workflows import TestWorkflow

workflow = TestWorkflow(
    test_command="pytest tests/",
    capture_screenshots=True,
    visual_regression_check=True
)

result = workflow.execute()

4. PR Creation with Artifacts

Problem: Creating PRs with screenshots/demos requires multiple manual steps

Solution: Single workflow from code change to PR with all artifacts

from openadapt_bootstrap.workflows import PRCreationWorkflow

workflow = PRCreationWorkflow(
    branch="feature/new-viewer",
    generate_screenshots=True,
    run_tests=True,
    create_demo=True
)

pr_url = workflow.execute()
print(f"PR created: {pr_url}")

Mobile-First Development

User on mobile can trigger workflows via:

1. GitHub Issues: Comment with workflow trigger

   /bootstrap run screenshot_workflow
   

2. GitHub Actions: Manually triggered workflows

   name: Generate Screenshots
   on: workflow_dispatch

3. Tailscale SSH: Direct command execution

   ssh desktop "cd ~/oa/src/openadapt-bootstrap && python -m workflows.screenshot_workflow"

Results posted back to GitHub for mobile review.

Integration with OpenAdapt Ecosystem

With openadapt-capture

from openadapt_capture import Recorder
from openadapt_bootstrap import WorkflowRecorder

# WorkflowRecorder wraps openadapt-capture Recorder
# with development-specific metadata and structure

With openadapt-evals

# Generate benchmark screenshots automatically
from openadapt_bootstrap.workflows import BenchmarkScreenshotWorkflow

workflow = BenchmarkScreenshotWorkflow(
    benchmark_dir="benchmark_results/my_eval_run",
    output_dir="screenshots/"
)

With openadapt-viewer

# Auto-generate viewer artifacts
from openadapt_bootstrap.workflows import ViewerWorkflow

workflow = ViewerWorkflow(
    capture_path="my_capture/",
    generate_html=True,
    generate_gif=True
)

Installation

cd /Users/abrichr/oa/src/openadapt-bootstrap
uv sync

Quick Start

Record a workflow

# Start recording
uv run python -m workflows.record --name my_workflow

# Perform task manually...

# Stop recording (Ctrl+C)

Replay a workflow

# Replay autonomously
uv run python -m workflows.replay --name my_workflow

# With Claude Code for decision-making
uv run python -m workflows.replay --name my_workflow --claude-code

List workflows

uv run python -m workflows.list

Proof of Concept: Screenshot Automation

See examples/generate_benchmark_screenshots.py for a working example that:

1. Launches benchmark viewer HTML
2. Uses OpenAdapt to navigate and capture screenshots
3. Saves to screenshots/ directory
4. Commits and pushes to PR branch
5. All without user intervention

uv run python examples/generate_benchmark_screenshots.py \
    --html-path benchmark_results/viewer.html \
    --output-dir screenshots/ \
    --commit-to-pr

Roadmap

Phase 1: Foundation (Current)

• Repository structure
• Architecture documentation
• Basic WorkflowRecorder implementation
• Basic WorkflowExecutor implementation
• Proof-of-concept screenshot workflow

Phase 2: Claude Code Integration

• ClaudeCodeWorkflow base class
• GitHub issue trigger system
• Result posting to GitHub
• Error handling and retry logic

Phase 3: Workflow Library

• Screenshot workflow (PR automation)
• Demo generation workflow
• Test execution workflow
• PR creation workflow
• Documentation update workflow

Phase 4: Self-Improvement

• Record workflow for "record workflow"
• Record workflow for "execute workflow"
• Fully recursive self-improvement loop
• Workflow optimization via evaluation

Related Projects

• openadapt-capture - Recording infrastructure
• openadapt-evals - Benchmark evaluation
• openadapt-viewer - Visualization

License

MIT

Contributing

See CONTRIBUTING.md