Add HTTP server API and run resumption support

[htyredc]

Summary

This PR adds a complete HTTP server API for managing Webwright exploration runs and introduces support for resuming prior runs with follow-up instructions. The server exposes run lifecycle management, execution of parameterized functions, and live progress streaming over Server-Sent Events.

Key Changes

Server API (src/webwright/server/)

• app.py: FastAPI application exposing HTTP endpoints for run management, follow-ups, script execution, and health checks
• run_manager.py: Single-active-operation manager ensuring only one run/follow-up/execution can be live at a time; handles process lifecycle, reconciliation after server crashes, and re-attachment to orphaned workers
• worker.py: Subprocess entrypoint for exploration runs and follow-ups; runs outside asyncio event loop to avoid nested-loop conflicts
• replay_worker.py: Subprocess entrypoint for executing parameterized functions or replaying recorded steps against a live browser session
• artifacts.py: Read-only helpers for accessing per-run artifact directories (trajectory.json, debug steps, final_function.py, etc.)
• schemas.py: Pydantic models for HTTP request/response validation
• sse.py: Server-Sent Events stream for live run progress by polling filesystem artifacts
• config_presets.py: Built-in YAML config preset enumeration and server defaults
• doctor_api.py: JSON adapter over environment health checks

Run Resumption

• src/webwright/utils/trajectory.py: Utilities to load trajectory.json, extract resume state, and resolve trajectory paths
• src/webwright/run/cli.py: Extended run_one() to accept resume_from and followup parameters; integrates resume state into agent initialization
• src/webwright/agents/default.py: Added resume() method to agents for continuing from prior state; drops exit messages and appends follow-up instructions

Environment Updates

• src/webwright/environments/local_browser.py: Added target-finding helpers (_find_cdp_target_by_id, _find_cdp_target_by_url, _find_new_cdp_target) and resume configuration options (resume_existing_output, resume_existing_page)
• src/webwright/environments/local_workspace.py: Added resume_existing_output config option

Configuration

• src/webwright/config/crafted_browser_fn.yaml: New config preset for authoring reusable parameterized browser functions (mirrors crafted_cli.yaml for live browser context)
• src/webwright/config/model_claude.yaml: Updated model endpoint configuration

Testing

• tests/unit/test_server_run_manager.py: Tests for run manager state, operation queueing, and process lifecycle
• tests/unit/test_server_app.py: Tests for HTTP API endpoints and run CRUD operations
• tests/unit/test_server_artifacts.py: Tests for artifact parsing (params spec extraction, trajectory info)
• tests/unit/test_resume_flow.py: Tests for trajectory loading, resume state building, and agent resumption
• tests/unit/test_model_json_parsing.py: Tests for JSON output parsing with various formatting styles

Documentation & Dependencies

• README.md: Updated installation and usage instructions to use uv package manager; added resume workflow examples
• pyproject.toml: Added [server] optional dependency group (fastapi, uvicorn); added dev dependencies for testing
• src/webwright/run/doctor.py: Generalized API key check to support multiple model backends (OpenAI, Anthropic, OpenRouter)

Notable Implementation Details

• Process Management: Worker processes run in separate process groups to enable unconditional cancellation via SIGTERM; re-attachment after server restart is supported via PID liveness checks
• Filesystem-Based IPC: Server and workers communicate through JSON files (server_meta.json, server_request.json) rather than pipes, enabling clean

https://claude.ai/code/session_01S3APtC5izAUzVHk2W3qsdE