Repository navigation
Conversation
Size Report
Startup median (7 runs, lower is better):
|
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
All reported issues were addressed across 53 files
Reply with feedback, questions, or to request a fix.
View guided diff | Turn on auto-fix | Re-trigger cubic
|
The rewrite keeps the meaning in the hunks I checked, and one Cubic thread still needs a fix at 9c7d260. The Not blocking: These Cubic threads do not apply, so please resolve them. The base text had the same gap or the same meaning in each case:
I did not compare all 53 files sentence by sentence, so meaning drift outside those threads and the hunks I sampled (workflow, metro reload, settle, fold, open) is not verified. I also did not re-run the |
|
One more thing: main moved after my review, and 9c7d260 now has merge conflicts with main. Please rebase and resolve them together with the findings above. |
|
Thanks for the update. The earlier Smoke Tests failure does not recur: the six checks at 9245e80 show no failures, and the change since the last review only touches a bench script that CI does not run. One finding from the 9c7d260 review is still open, and the PR has merge conflicts with main. The Not blocking: the change to On the open inline threads, the
I reviewed the code at 9245e80 and found one remaining defect. Before merge, please restore the iOS-simulator scope on the I could not verify that the gpt-5.6-luna model exists or that the 111/186 bench parity holds, since both come only from the PR body. I did not re-check meaning drift sentence by sentence outside the new delta, because the earlier review sampled hunks and commits 1-65 are unchanged in the range-diff. I also did not re-run the |
…ves inline The debugging, tv, react-native, remote, and macos routes stay in the workflow topic as inline pointers. The deletion makes room for the ASD-STE100 rewrite inside the 9,100-byte workflow budget.
The workflow and manual-qa topics listed themselves in the shared Related footer. The self-link repeats the topic that the reader is on.
Restore the routes that the deleted Escalate entries gave: - alerts: help debugging - RN hazards and routing: help react-native - remote config, leases, tunnels: help remote - desktop, frontmost-app, menu bar: help macos Delete the duplicate 'snapshot -i is for refs.' sentence to keep the topic below the 9,100-byte budget.
callHelpTool returns buildUsageText() when the topic is absent, and buildUsageText() renders renderCliHelpOverview(), the root decision card. The description said that it returns the full workflow card.
launchConsole said 'Launch console mode.', but the value is a file path: - packages/platform-apple/src/core/app-launch.ts runIosSimulatorConsoleLaunch calls ensureHostDirectory(path.dirname(logPath)) and writes the launch stdout/stderr to that path (simctl launch --console-pty). - openIosApp and assertOpenDeviceSupport refuse it off an iOS simulator (LAUNCH_CONSOLE_IOS_SIMULATOR_ONLY_MESSAGE). - invokeApplicationUrlOpen and invokeApplicationTargetOpen refuse it for URL opens (LAUNCH_CONSOLE_DIRECT_APP_ONLY_MESSAGE). - The --launch-console flag text already says 'open: capture the initial iOS simulator launch console window to a file'. jitterPx said 'Randomization radius in pixels.', but the offsets are deterministic: packages/contracts/src/touch-runtime.ts pressJitter(index, amount) cycles the fixed PRESS_JITTER table by press index and scales it by amount. The --jitter-px flag text already says 'deterministic coordinate jitter'.
Base named the platform scope of help tv (Android TV, tvOS, Vega VVD remote). Put the scope back inline in the Snapshots and refs section. Delete the Escalate entry for help scripting: the Validation and evidence section gives the same route with the same scope, so the topic stays inside the 9,100-byte budget.
…vice first screen Base line 26 told the Mac DevToolsSecurity reason apart from the device Developer Mode reason, gave the device-state gating conditions, and said that the two device reasons are never inferred. The rewrite moved these rules below line 30. Put compact versions back on line 26 and move the never-inferred rule there. The reason list below keeps the details.
The rest of the help uses examine for this concept. The macOS topic and the network command summary still used inspect. Keep "Do not write inspect as a command": there, inspect is the literal command word.
Agents often read only the first screen of a help text, and the help benchmark caps the root help at 30 lines. Several rewrites moved safety rules below that screen. Declare, for the root help and for every topic id, the key rules from the base rendering that must stay in lines 1-30: prerequisites, refusals, verification steps, and recovery. A topic id without an entry fails the test.
gpt-5.4-mini is not available when Codex runs with a ChatGPT login, so every Codex call in a default run failed. gpt-5.6-luna is the small model that the base and head runs for this PR used.
…e sentences Restores the platform qualifier that the base clause carried: - screenshot --pixel-density applies only to iOS simulators. - screenshot --fullscreen for the full desktop applies to macOS. - --settle-quiet: the tab-switch exception applies on iOS; split the long sentences. #3735 (comment) #3735 (comment)
…g sentences The base tied the logs-over-simctl rule to iOS simulators and the IME retry rule to the Android Gboard bullet.
… sentences - macOS: the XCTest-backend rule also scopes interactions that bring the app forward. - gestures: the metric rule applies to the iOS simulator transform/pinch/rotate gestures. - react-native: keep the worktree build reuse as a capability, not an instruction.
9245e80 to
3e7f1ef
Compare
|
Follow-up at 3e7f1ef:
Threads: 2 fixed (pixel-density, settle-quiet), 8 do not apply because the base text had the same meaning (resolved, as asked). None are rejected and open. CI: the earlier Smoke Tests failure did not recur at 9245e80. Checks for 3e7f1ef are pending. Validation: Not done: the |
|
The PR is ready. At 3e7f1ef, the follow-up changes fix the two points from the earlier review at 9245e80 (#3735 (comment)): the screenshot help now scopes Of the open inline threads, the two on #3735 (comment) and #3735 (comment) are fixed at this commit, so you can resolve them. The other eight do not need changes because the base text said the same: #3735 (comment) (the base airplane line also had no failure contract), #3735 (comment) (the base said the same), #3735 (comment) (keeps the base CI is green: 17 checks, none failing at 3e7f1ef. There are no conflicts. I did not re-audit commits 1-66 sentence by sentence for meaning drift, because they are unchanged in the range-diff. I did not check that the Android-scoped formatting rule in |
thymikee
added
the
ready-for-human
This branch has not been deployed
1 participant
Summary
Rewrites the versioned CLI help in ASD-STE100 Simplified Technical English: the root help, all help topics, the MCP guide, and the command, flag, and input metadata that help and MCP render. This is a language change; the meaning of each rule is frozen. Command lines, flags, enums, defaults, limits, and error codes are unchanged. One commit per topic or metadata module.
src/commands/schema/cli-help-first-screen.test.ts: each topic and the root help declare their key rules, which must stay in the first 30 rendered lines. A new topic must declare its rules. Detection is proved: a moved rule, a missing entry, and the earlier regressions all fail it.fix(help)commits correct facts: thelaunchConsoleandjitterPxinput descriptions, the MCP help-tool description ("root decision card"), and a topic that listed itself in its Related footer.Review rounds
Not changed (pre-existing)
Validation
pnpm check:affected --run --head 3e7f1efce6: passed,exactCommit=true.pnpm bench:help-conformance(31 cases,claude-haiku-4-5andgpt-5.6-luna, 3 repeats): base 111/186, head 111/186. A Codex drop on one case did not repeat in a parallel 10-repeat run (base 4/10, head 4/10). The default Codex runner is nowgpt-5.6-luna;gpt-5.4-minidoes not work with a ChatGPT login.