test(browser): verify semantic role and name before controlled action

[seonghobae]

Partial implementation of #28, stacked on PR #70.

Buyer-visible boundary

PR #70 proves real pinned-Chromium input/click/post-condition execution for the controlled Agent Task. This lane adds bounded browser-computed accessibility role/name evidence before either controlled action proceeds.

Reconstructed stack

The independently refetched live prerequisite is exact PR #70 head 4930ab764eb63c524c4129196bd2acb6188c1c02. The prior #71 lineage had merge base d1f0dfc611c07b847477bfe4671eba0a4d83d065 and had fallen behind the moved prerequisite.

This lane was reconstructed non-destructively from the current #70 tree while preserving #71's semantic-evidence code/tests/docs and reapplying its unique CHANGELOG statement onto the current prerequisite rather than carrying a stale whole-file blob. Current exact head is 94c962463c2adebf3e8510057cf9c948ee3fba18.

Fresh comparison is ahead 16 / behind 0, with merge base exactly current #70. The live-base delta is limited to CHANGELOG.md, docs/DOCUMENTATION_FITNESS.md, docs/TEST_STRATEGY.md, docs/traceability/action-postcondition-evidence.md, scripts/ci/run_mv3_compatibility.py, and tests/test_agent_task_pinned_chrome_contract.py. No prerequisite/predecessor check or review evidence transfers.

Implemented boundary

The runner obtains browser-computed W3C WebDriver role and accessible-name values before the controlled clear/type or click action. It requires exact textbox / Task text and button / Submit task semantics, fails closed on malformed or mismatched semantics, and preserves #70's same-document URL invariant, synthetic input echo, post-condition proof, extension isolation, repeatability, and profile cleanup.

The current automated review's three inline notes were informational verification only: the role/name expectations match the fixture, semantic reads occur safely before input mutation, and the documentation contract substrings are present. They identified no source defect; those addressed informational threads were resolved without code churn.

Exact-current evidence

On unchanged exact head 94c962463c2adebf3e8510057cf9c948ee3fba18 against exact prerequisite 4930ab764eb63c524c4129196bd2acb6188c1c02:

• ancestry: ahead 16 / behind 0, merge base exactly the live prerequisite;
• CI run 32957590483: success;
• Rust contracts job 98142678641: repository contracts, canonical formatting, locked workspace/all-target check, full tests, strict Clippy, and rustdoc success on exact checkout;
• Production coverage job 98142678314: exact owned-production function/line/region/branch measurement and enforcement success;
• exact owned production coverage: functions 433/433, lines 3675/3675, regions 4580/4580, branches 512/512 — all 100%;
• exact coverage artifact 9602632987, digest sha256:955ce8ad0eb1de8425042ab1e7c48396a335c23e3c2f65078e1b098329b5ba86;
• Manifest V3 Compatibility run 32957590541: success;
• pinned-Chrome MV3 job 98142606538: success on exact checkout;
• MV3 artifact 9602581517, digest sha256:c86ffbeb10d5141d0bb13428691213d8eb8e970cb6f872a3d706772095a52c92;
• formal review inventory contains automated COMMENTED reviews only and no qualifying independent approval; and
• current inline review threads are resolved.

Same-head SAST/Semgrep, Security Scan, OpenCode, Noema, Strix, or qualifying approval evidence not returned for this feature-base head is absent, not passing. No predecessor, prerequisite, queued, skipped, cancelled, synthetic, status-only, or model-only result is promoted.

Truth / authority boundary

This is executable controlled-browser evidence, not the final OriginWeave browser runtime. CSS selectors remain controlled fixture locators; computed role/name proof does not itself implement a product semantic observer, opaque node-handle authority, policy dispatch, provenance, trusted process attribution, or real-site compatibility. PR #70/#65 remain active prerequisites.

Protected-main AGENTS.md remains authoritative. This scheduled actor does not merge, self-approve, force-push, destructively rebase, alter workflows/rulesets/secrets, weaken checks, tag, release, or publish.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 9dd1464b-4f3d-470a-a395-b0f467b53ab9

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[devin-ai-integration]

Devin Review found 3 new potential issues.