Repository navigation
docs(skill): route agent-driven browser actions to scenario capture - #2286
Conversation
|
Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits. |
|
@greptileai review |
|
This PR is set for a Plus review. TREX isn't supported with Plus or Apex reviews yet, so Greptile reviews this PR at Base while TREX is on. |
|
|
Review findings addressed at their producing boundaries and consumers. Latest changes were reviewed from source; tests and provider runs were skipped at the maintainer's request. Findings requiring separate decisions or verification remain open. Fixed in Validation: inspected the producing boundary and affected consumers; existing expectations were adjusted where the contract changed. Tests, mutation runs, builds, lint, typecheck and CI reruns were skipped at the user's explicit request. No new real-provider verification is claimed. This source correction does not settle separate integration authority or mutation-generation/score proof requirements. |
Integrate the verified main revision without rewriting the contributor's history. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Distinguish final capture artifacts from step outcomes, bound listener attribution, and request only the evidence needed for the analyst question. Update the bundled skill version and its generated package metadata together. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Integrate the verified main revision without rewriting the contributor's history. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
Distinguish final capture artifacts from step outcomes, bound listener attribution, and request only the evidence needed for the analyst question. Update the bundled skill version and its generated package metadata together. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
N0zoM1z0
left a comment
There was a problem hiding this comment.
Thanks for adding this route. Reviewed the final instructions against the current browser contracts and verified real launch/connect CLI scenarios, requested artifacts, initial navigation and cleanup on Linux Chromium. Documentation, typecheck/lint, formatting, generated-skill tests and package verification passed. The initiator and listener limitations are now explicit; the skill version and generated metadata agree.
|
Want your agent to iterate on Greptile's feedback? Try greploops. |
N0zoM1z0
left a comment
There was a problem hiding this comment.
Confirmed the header wording issue in safeResponseMetadata and its boundary test: selected content encoding, policy values and Link URLs remain in passive evidence. I have stopped the merge watch and am correcting that paragraph with a real CLI metadata probe.
Distinguish discarded raw header collections from the selected response projection, which preserves encoding, policy values and Link URLs. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
N0zoM1z0
left a comment
There was a problem hiding this comment.
The header projection wording is corrected and matches both the production producer and a real public CLI capture. The passive metadata probe and 44 focused tests passed, as did updated documentation and package verification. Previously verified scenario routes, ownership, artifact selection and initiator limits remain unchanged.
N0zoM1z0
left a comment
There was a problem hiding this comment.
The header projection wording is corrected and matches both the production producer and a real public CLI capture. The passive metadata probe and 44 focused tests passed, as did updated documentation and package verification. Previously verified scenario routes, ownership, artifact selection and initiator limits remain unchanged.
Integrate the verified main revision without rewriting the contributor's history. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
N0zoM1z0
left a comment
There was a problem hiding this comment.
Reviewed the main synchronization. The PR still changes only the two authored skill files and their derived version metadata. The inherited npm cache-policy failure is fixed by main #3013; all 54 focused tests, documentation checks, typecheck/lint, formatting and full package verification passed on this head. Browser producer/contracts are unchanged from the real CLI probes.
|
Both previous CI failures have been repaired on main. Synced those fixes in |
Bring in the main fixes for the Ghidra profile fixture and isolated CI report tests. Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com> Co-authored-by: factory-droid[bot] <138933559+factory-droid[bot]@users.noreply.github.com>
N0zoM1z0
left a comment
There was a problem hiding this comment.
Reviewed the synchronized revision: the PR still changes only two skill documents and matching version metadata. Local docs, typecheck/lint, formatting, 61 focused tests and real Chrome launch/connect/passive CLI probes passed. Both previous CI failures were repaired on main and included in this revision.
Summary
Choose browser tools by who performs the action: passive observation for user actions, declared
capture_browser_scenarioactions for agent tasks.The skill explains connect/launch ownership, initial navigation and unsaved state, requested capture details, separate step outcomes, and request-initiator limits. Listener locations identify candidate handlers. It also lists focused follow-up tools and preserves passive-only reconciliation guidance.
Bump the bundled skill from 41 to 42 and regenerate package metadata. The diff contains two authored skill files and that derived version field.
Synced main through
e049ef70to include the repaired Ghidra profile fixtures and isolated CI report tests.Validation
On Linux with pinned Node.js 24.18.0/npm 11.16.0:
docs:check: five conformance tests passed.check:fast, full formatting and diff checks passed.verify:packageon738d02a: repository skill installation, package install/update, CLI/MCP workflows and cleanup passed. It was not rerun after this main sync.An additional real
inspect-web-pageCLI probe retained gzip encoding, COOP policy and a Link query value while omitting credential/unselected header values.Separate follow-up browser tools were checked against source contracts. Native Windows/macOS and agent transcripts were not rerun. Browser output sizes are observations, not token or performance benchmarks.
Contributor's earlier macOS evidence
The contributor reports checks on REA 6.4.0/main
d96550f, macOS 26.7.1, Chrome 154 and Brave 155. Connect scenarios clicked#export, retained a 200 response and the expected final DOM, then disconnected. Listener inspection locatedexport.js; the request initiator came from passive capture. Overlapping passive capture and scenario navigation returnedtarget_unavailable/target_changed. Their connect capture remained incomplete due to pre-attach event coverage.