Fix typo in readme

merge

Fix/codeceptjs is undefined

…atched typing

Attacks the fixed-overhead costs identified in the maintainer's Obscura-vs-headed-Chrome
benchmark: amOnPage/refreshPage now race a push-based Page.lifecycleEvent 'load' signal
(matched by loaderId, correctly discriminating a navigation from stale/about:blank lifecycle
events) against the readyState poll, so engines that emit it skip most of the poll's latency;
the poll's own retry interval is decoupled from waitForAction into a new, lower pollInterval
option (25ms default) so waitForAction stays untouched for click/type pacing parity. _runSelected
folds the client-presence check into the same round-trip as the action itself via a sentinel
value, only paying a separate _ensureClient round-trip when the client is actually missing.
type() dispatches every character in one round-trip when no delay is requested, keeping the
per-character server-side loop only when delay > 0 needs real pacing.

Mandate suspect #2 (persistently registering the client via
Page.addScriptToEvaluateOnNewDocument, to avoid re-injecting ~180KB per navigation) was
implemented and measured, then reverted: Obscura accumulates internal state per registration
call across many Target lifecycles within one CDP connection, driving Page.navigate's ack time
from ~8ms to 20-30ms+ within 40 iterations and, in one isolated reproduction, into an outright
10s command timeout. The manual, reactive injection path (unchanged) is faster and safe.

Fix 5 (auditing waitForAction call sites against Playwright's parity default) found nothing to
change: Playwright's actual code applies the same pacing to a superset of CDPBrowser's 11 call
sites (its JSDoc naming only click/doubleClick/pressKey is stale documentation), so there is
nothing to trim without diverging from parity.

Full before/after profiles, the fix-2 rejection evidence, and suite timings are recorded in
.superpowers/skip-sweep-report.md.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>