Communitygithub.com

AlphaZero78/browser-question-workflow

A reusable Codex Skill for reliably completing and verifying authorized browser question workflows.

browser-question-workflow 是什么?

browser-question-workflow is a Codex agent skill that a reusable Codex Skill for reliably completing and verifying authorized browser question workflows.

兼容平台~Claude CodeCodex CLI~Cursor
npx skills add AlphaZero78/browser-question-workflow

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

Browser Question Workflow

Operate the user's existing browser session and finish authorized question workflows end to end.

This workflow is developed and verified primarily for U校园 / Unipus. Treat all other websites as unverified adaptations: inspect each interaction and never assume that submission or completion detection is correct.

Rules

  • Use answers supplied by the user, a user-designated answer file, or answers explicitly revealed by the page.
  • Do not infer answers for graded assessments. Ask for the missing answer source or leave those questions unanswered.
  • Preserve the logged-in session. Prefer Chrome control for existing Chrome state.
  • Treat a click on 提交 as an action requiring explicit or already-given user authorization.
  • Skip question types the user excluded, such as speaking or recording tasks.
  • Never equate "visited", "answered", or "submitted one page" with "section completed".
  • Outside U校园 / Unipus, use short steps and verify every write, navigation, submission, and completion signal before continuing.

Workflow

  1. Inventory the task

    • Identify target units, sections, exclusions, answer sources, and whether submission is authorized.
    • Convert answers into a structured map keyed by section and question number.
    • For randomized options, store exact option text, not only letters.
  2. Inspect live state

    • Read URL, visible buttons, tabs, question count, input count, dialogs, iframes, and completion icons.
    • Record which sections are complete, in progress, or untouched.
    • Start from the first incomplete allowed section.
  3. Determine page shape

    • Classify the page as one-question-per-page, all-questions-on-one-page, text blanks, mixed form, or cross-origin iframe.
    • Check whether option order changes between attempts.
    • Read references/browser-patterns.md for robust interaction patterns.
  4. Fill conservatively

    • Match radio answers by exact normalized option text when possible.
    • Use letters or indexes only after confirming the current option order.
    • Prefer bulk form filling for stable all-on-one pages.
    • For reactive text fields, dispatch realistic input and change events.
    • On one-question-per-page flows, answer, click 下一题, wait for navigation, then re-query the DOM.
  5. Submit and verify every section

    • Submit only after checking answered count against expected count.
    • Wait for a result, 继续学习, score, review state, or completion marker.
    • After moving to the next section, verify the previous section's directory icon changed to completed.
    • If an item remains "in progress", inspect its child pages before assuming the cause.
  6. Handle tests and iframes

    • Click entry confirmation, start the test, then wait for resources to finish loading.
    • Confirm and dismiss any "resources loaded" overlay before filling.
    • Use accessibility snapshots and form-fill tools for cross-origin iframe contents.
    • Submit, verify score/answered count, then use the test system's return or course link so the parent course receives the result.
  7. Recover from interruption

    • Re-inspect completion icons and current question state.
    • Re-enter only incomplete sections.
    • Refill a section when the page shows answers but no completed marker.
    • Keep a concise progress ledger: completed, skipped by instruction, pending, and blocked.
  8. Final audit

    • Re-read all target completion icons after navigation or refresh.
    • Report completed sections, skipped question types, scores, and any remaining in-progress reason.

Efficiency

  • Use one inspection call to collect URL, visible buttons, links, input counts, and completion markers.
  • Batch independent DOM reads; batch stable form writes.
  • Use direct section links or known routes only after confirming they belong to the current course.
  • Wait after navigation and submission; do not run long scripts across page transitions.
  • Prefer short, restartable operations over one fragile script covering many pages.

Completion Standard

Call a section complete only when at least one durable signal confirms it:

  • the directory marker says completed;
  • a result page shows the expected answered count or passing score;
  • the page exposes a submitted/review state and the course records it.

For sections containing excluded speaking/recording tasks, report the non-excluded portion as submitted and explain why the parent section remains in progress.

相关技能