scirepl-playwright

SciREPL Playwright workflow

Use this skill when a task requires running, inspecting, or troubleshooting a SciREPL workbook through the Playwright driver.

Session rule

Complete one workflow in one scirepl_* browser session. A separate general browser or Playwright MCP normally has different storage and in-memory state. Only treat another client as the same browser when it explicitly attaches to the CDP URL with connectOverCDP.

Standard workflow

1. scirepl_connect
2. scirepl_install_package or scirepl_import_file, when needed
3. scirepl_run_all_cells_ui for visible testing, or scirepl_run_all_cells for batch execution
4. scirepl_get_cell_outputs_detailed
5. scirepl_get_visible_state and scirepl_screenshot for supporting evidence
6. scirepl_disconnect

Prefer scirepl_run_all_cells_ui when the user mentions the menu, visible output, warnings, browser behaviour, or screenshots.

Diagnosing output

1. Reproduce the issue through the requested run path.
2. Read scirepl_get_cell_outputs_detailed; it combines app-model output with DOM-rendered text and includes Out [n] labels where present.
3. Capture a screenshot for layout and rendering evidence.
4. Use console messages from scirepl_get_visible_state as supporting evidence, not as a substitute for the visible cell result.
5. If the user sees output that the tool omits, treat that as an instrumentation gap and inspect the driver before concluding the output is absent.

Browser-state investigation

For stale assets, cache, storage, workers, or PWA-update issues:

1. scirepl_connect(debugMode: true, remoteDebuggingPort: 9223)
2. scirepl_get_browser_debug_info
3. Continue the ordinary workflow in that same session
4. If necessary, attach a CDP-capable debugger to the reported URL

Opening the debugger URL as a web page is not equivalent to attaching through CDP.

Supported kernels

Python, R, TypR, SWI-Prolog, Bash, JavaScript, Lua, and ClojureScript.

Local package iteration