Skip to content

v1.5.0 — the validation-layer release

Choose a tag to compare

View all tags
@PramodDutta PramodDutta released this 07 Jul 15:12
v1.5.0
845a0c4

BrowserBash becomes directly consumable by AI agents, and verdicts stop being model judgment.

Highlights

  • MCP server: browserbash mcp — plug into Claude Code, Cursor, Codex or any MCP host with one line: claude mcp add browserbash -- browserbash mcp. Tools: run_objective, run_test_file, run_suite.
  • Deterministic assertions: Verify ... steps compile to real Playwright checks (no LLM judgment); run_end.assertions carries expected-vs-actual evidence.
  • testmd v2: version: 2 frontmatter — per-step execution on one browser session, plus deterministic API steps for arrange-act-assert suites.
  • Saved logins: browserbash auth save <name> + --auth — log in once, reuse everywhere.
  • Monitor mode: browserbash monitor --every 10m --notify <webhook> — alerts only on pass/fail changes; Slack autodetected.
  • Cost governance: cost_usd estimates, suite --budget-usd hard stop.
  • Sharding + viewport matrix: --shard 2/4, --matrix-viewport.
  • Playwright import: browserbash import ./e2e with an honest IMPORT-REPORT.md.
  • Flow recorder: browserbash record <url> — passwords never captured.
  • GitHub Action: action.yml at repo root; PR verdict comments; see docs/github-action.md.

Full details in CHANGELOG.md. 187 tests. Exit codes and NDJSON schema unchanged (additive only).

Assets 2
Loading