Reuse fixes across runs¶
Within a run, heal remembers each broken→healed locator and swaps it proactively on the next occurrence (no second LLM call). Warm start extends that across runs by loading known fixes from the healing history.
How it works¶
At the first qualifying failure, heal loads recent healed mappings (scoped per
source file) from history.sqlite. When a known-broken locator appears again and
still matches nothing, heal swaps it before the keyword runs — skipping both
the keyword timeout and the LLM entirely. Every swap is verified live
(broken matches 0, healed matches 1), so a stale mapping simply falls through to
normal healing.
run 1: Click id=login-button → heal → css=#signin-btn (LLM, ~5s)
run 2: Click id=login-button → warm swap → css=#signin-btn (0 tokens, ~0s)
Enable a stable history location¶
Warm start is on by default (HEAL_WARM_START=true), but the default history
lives under the run's output directory — which changes per run. Point it at a
stable path so it survives:
Now a second CI run heals known breakages with zero tokens. Reused swaps are
recorded as events (marked reused-from-history) so they still appear in the
report and summary.json.
Inspect repeat offenders¶
Locators healed repeatedly are maintenance hotspots — fix them permanently with Fix test files.