Encourage accessibility snapshots in Playwright results - #169
Merged
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
rgarcia
changed the title
rgarcia
force-pushed
the
hypeship/clawbench-aria-snapshot
branch
from
August 28, 2026 23:34
ace5e7a to
54ca49b
Compare
rgarcia
force-pushed
the
hypeship/clawbench-aria-snapshot
branch
from
August 29, 2026 00:35
54ca49b to
a2bcbcc
Compare
rgarcia
force-pushed
the
hypeship/clawbench-aria-snapshot
branch
2 times, most recently
from
August 30, 2026 18:23
474b3b3 to
2466301
Compare
rgarcia
force-pushed
the
hypeship/clawbench-aria-snapshot
branch
from
August 31, 2026 15:16
2466301 to
edfec5a
Compare
ClawBench · PR #169 (edfec5a) vs merge base (e96298a)
Candidate minus baseline: -2 lenient, +2 strict, -1 intercepted. Open the Braintrust experiment Lenient reward is the primary ClawBench score. Infrastructure failures remain in the intended-task denominator. |
Contributor
Author
|
@bmsaadat the full 129-task candidate-vs-base run completed on the latest head. Candidate versus baseline: -2 lenient, +2 strict, and -1 intercepted; all graded trials passed exact Kernel MCP session validation. The benchmark comment links the 258-row Braintrust experiment. Requesting review for the one-line tool-description change. |
bmsaadat
approved these changes
Aug 31, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
execute_playwright_codeinput description to ask every invocation for useful page stateariaSnapshot()after navigation and interactionWhy
Browser agents need the result of each Playwright call to decide the next action. Explicitly returning a compact accessibility snapshot keeps that state useful without producing a full-page payload.
This is intentionally separated from #162 so the ClawBench workflow can benchmark this description-only change against its base SHA.
Replay
Watch the successful Kernel MCP ClawBench replay for
v2-1202-beauty-cart-quantity-theordinaryat 4× speed.The task requires the agent to find Multi-Peptide + HA Serum, open the correct product, change its quantity to
2, and add it to the cart. ClawBench intercepted the resultingCart-AddProductrequest and the strict judge verified that it contained the correct product and quantity. The run scored1.0for interception and strict reward.This run exposed only Kernel MCP
execute_playwright_codeto Codex with GPT-5.6 Luna. Its source was the exact description-change commit37273e3, and the trajectory usedariaSnapshot()in 11 browser calls. The published 51-second replay contains no credentials, signed Kernel URLs, session IDs, or private benchmark artifacts.Test plan
bun test src/lib/mcp/register.test.tsbunx tsc --noEmitbunx prettier --check src/lib/mcp/tools/playwright.ts