Help me correct my email address and password on the signup form.
The intention is included in the agent’s input prompt alongside environment details and task instructions.
BeforeThe defect at the parent commit
0:00.00 / 0:00.00Frame 0 · Replace short in Password with Kestrel!482
Watch the password correction and compare the entered value with its minimum-length validation message.
AfterThe same script once the merged fix is applied
0:00.00 / 0:00.00Frame 0
Correct the email and password in turn and inspect the validation feedback beside each field.
01 · Given
Restore the data and starting state.
On the signup form, before submission, Full Name is Alex Morgan, Email Address is alex, and the revealed Password is short. Email Address and Password have both been blurred and show their respective invalid-email and minimum-eight-character errors.
02 · When
Perform the interaction.
Append @example.com to Email Address while it retains focus, then move focus to Password and replace short with Kestrel!482 while retaining focus in Password. Do not submit the form.
03 · Judge
Rationale: Signup users need to know whether correcting their credentials worked. An obsolete error makes a valid value appear invalid and obstructs recovery even though the input now satisfies the rules.
Require
After an email or password has shown a validation error, valid correction must promptly clear that error and its invalid styling while the field remains focused, without blur or submission. Invalid values must retain accurate validation feedback.
Allowed
- Revalidation may occur during input or after a short typing pause, provided feedback updates without a focus change or submission.
- Success feedback may be omitted or shown in any readable form; exact wording, colors, layout, and the validation library are unrestricted.
- Initial validation may wait until blur; this case requires timely revalidation of errors already shown, not an error on the first keystroke.
Model
HarnessGPT-6 Astra
Codex
815.8k in · 3.9k out · 819.7k total
Codex · 28 steps · $1.35 · 10m 42s · 819.7k tokens
Codex
28
$1.35
10m 42s
815.8k
3.9k
819.7k
DeepSeek Flash
DSH
9.4M in · 38.4k out · 9.4M total
DSH · 86 steps · $0.07 · 15m 32s · 9.4M tokens
DSH
86
$0.07
15m 32s
9.4M
38.4k
9.4M
GLM-5.3 Flash
ZCode
15.9M in · 64.0k out · 15.9M total
ZCode · 176 steps · $0.53 · 36m 46s · 15.9M tokens
ZCode
176
$0.53
36m 46s
15.9M
64.0k
15.9M
Claude Opus 5.5
Claude Code
573.3k in · 7.6k out · 580.9k total
Claude Code · 18 steps · $0.54 · 8m 58s · 580.9k tokens
Claude Code
18
$0.54
8m 58s
573.3k
7.6k
580.9k
Kimi K3
Kimi Code
7.9M in · 88.2k out · 8.0M total
Kimi Code · 73 steps · $4.24 · 1h 05m · 8.0M tokens
Kimi Code
73
$4.24
1h 05m
7.9M
88.2k
8.0M
Qwen3.8 Max
Codex
15.4M in · 96.2k out · 15.5M total
Codex · 139 steps · $4.86 · 1h 05m · 15.5M tokens
Codex
139
$4.86
1h 05m
15.4M
96.2k
15.5M