flakewatch-core

Core logic for Flakewatch -- AI-powered Playwright test failure triage.

This package contains the shared libraries used by the CLI, reporter, browser agent, and GitHub Action. Most users should use flakewatch (CLI) or flakewatch-playwright (reporter) instead of this package directly.

What's inside

• Config -- loading, validation, and defaults
• Context -- packaging test failures with source code, page objects, error info, and screenshots
• Grouping -- clustering failures by normalized error signature
• LLM -- Anthropic client with verdict-focused system prompt
• Verdicts -- type definitions for all verdict types
• Reporting -- Markdown, JSON, and GitHub comment output
• Triage -- the main pipeline that groups, investigates, and reports

Usage

import {
  loadConfig,
  triageFailures,
  formatMarkdownReport,
  parsePlaywrightJsonReport,
} from 'flakewatch-core';

const config = await loadConfig();
const { failures, totalTests } = parsePlaywrightJsonReport(jsonReport);

const report = await triageFailures({ failures, totalTests, config });
console.log(formatMarkdownReport(report));

License

MIT