Making heals permanent — apply-heals

Runtime healing never touches your source. The original broken locator stays in your test file or Page Object forever, healed at runtime on every run, until you fix it. apply-heals closes that loop: it rewrites the broken locator to the selector that actually worked.

npx playwright test                          # heals at runtime, records to heals.jsonl
npx tamash-playwright apply-heals --dry-run  # preview the source changes
npx tamash-playwright apply-heals            # write them

Every run — dry or real — prints a table of exactly what it found before touching anything:

Fixes (1)
  ┌───────────────────────────┬──────────────────────────┬─────────────────────────────────────────────┬────────┐
  │ Location                  │ Before                   │ After                                       │ Review │
  ├───────────────────────────┼──────────────────────────┼─────────────────────────────────────────────┼────────┤
  │ src/pages/loginpage.ts:11 │ .locator('input[name="…' │ .getByRole("textbox", { name: "Username" }) │ —      │
  └───────────────────────────┴──────────────────────────┴─────────────────────────────────────────────┴────────┘

1 fix(es) applied to 1 file(s), 0 skipped.

Proving the fix stands on its own

Every real run also writes a verification script — .tamash-playwright/verify-heals.cjs (TS) or .tamash-playwright/verify_heals.py (Python):

node .tamash-playwright/verify-heals.cjs    # TS
python .tamash-playwright/verify_heals.py   # Python

This re-runs exactly the affected tests with HEALER_ENABLED=false. A pass proves the rewritten selectors work standalone — not just "worked while healing was still propping them up." In CI, this verification step runs automatically and its result goes into the PR body.

Good to know

• Nothing runs automatically. apply-heals is a separate, deliberate command — a test run never edits source on its own.
• Only real selector fixes qualify. Text / ARIA-based heals, including vision heals that resolved to something reusable. Never an action-recovery heal.
• Surgical edits. .describe('...') and everything else on the line is untouched — only the .locator(...) / .getByRole(...) call is replaced.
• needsReview is not a gate. A flagged fix is still applied; the flag just shows in the table and the reports.
• A real run asks for confirmation — at a real terminal only. In CI / non-interactive it proceeds (no stdin). --yes skips the prompt at a terminal too.
• Every run leaves a report — .tamash-playwright/apply-heals-report.md / .json, archived under history/.
• Always review before committing. It rewrites source files. git diff, re-run, commit deliberately.

In CI

Run it as a job after your test job, feeding it the heal-log artifact. --logs-dir <path> merges heals.jsonl from multiple sharded jobs. It opens a PR with the verified diff rather than committing directly. See Running in CI.