Treat college application planning as a decision process of “understand yourself → understand careers/majors/industries → constrain choices with admissions data → generate an auditable report.” The main interaction takes place in the Agent conversation; HTML is used only to present the final report and is no longer the primary data-entry interface.

At startup, first tell the user: This service is for reference only when making Gaokao college application decisions. All personality analyses, career recommendations, industry assessments, and admission estimates are based on public sources and model inferences; they cannot replace the provincial education examination authority, the university’s admissions prospectus, the official application system, or input from teachers or parents. Final choices must be cross-checked. The same notice must appear at both the top and bottom of the final HTML report.

The legacy and / / / are legacy compatibility assets for the web task queue workbench. The new main flow must not require the user to open a web page for data entry or make the waiter a mandatory step.

Use the script to generate a static HTML form from assets/personality-assessment-form.html and give it to the user to complete in a browser:

python3 <skill-dir>/scripts/create_assessment.py \
  --workspace <candidate workspace absolute path> \
  --open

automatically opens the form in the system’s default browser, so the user does not have to find it in a folder. The script outputs a file path such as by default. Tell the user the file path and explain: Please open the HTML form in your browser, and after completing it, click “Download JSON for Agent,” then send the downloaded JSON file back, paste its contents, or authorize the Agent to read the file. Do not ask the user to complete the questions one by one in the terminal chat.

The Markdown template assets/personality-assessment-template.md is only a fallback: if the user cannot open the HTML, run to generate a fillable Markdown form.

The questionnaire is based on the RIASEC Career Interest Inventory and the Mini-IPIP abbreviated Big Five personality inventory, with 32 items scored from 1 to 5.

Parse the personality assessment result JSON returned by the user (or the fallback Markdown), and score it according to the rules in the template. After scoring, first generate a readable HTML report so the user can confirm that they understand it:

python3 <skill-dir>/scripts/render_personality_report.py \
  --input <personality-assessment-result.json> \
  --output <candidate workspace absolute path>/personality-report.html

The report should present plain-language key conclusions first, followed by the full score table and limitations in expanded form.

The personality test can only serve as a directional clue; do not present its results as a diagnosis or a verdict on the person’s fate.

Generate 3–6 major clusters around the career direction hypotheses. Every major cluster must be researched for the following:

Preferred sources include the Ministry of Education’s major catalog, Sunshine Gaokao, official university websites for curriculum plans/admissions prospectus/employment quality reports, the National Bureau of Statistics, human resources and social security departments, industry regulators, and authoritative industry associations. Job boards and commercial reports may be used only as supplementary signals, and any sample bias must be stated.

Use the script to generate a static HTML form from assets/admission-info-form.html and give it to the user to complete in a browser:

python3 <skill-dir>/scripts/create_admission_info.py \
  --workspace <candidate workspace absolute path> \
  --open

automatically opens the form in the system’s default browser, so the user does not have to find it in a folder. The script outputs a file path such as by default. Tell the user the file path and explain: Please open the HTML form in your browser, and after completing it, click “Download JSON for Agent,” then send the downloaded JSON file back, paste its contents, or authorize the Agent to read the file. Do not ask the user to complete the fields one by one in the terminal chat.

The Markdown template assets/admission-info-template.md is only a fallback: if the user cannot open the HTML, run to generate a fillable Markdown form.

The Gaokao information document must cover the following:

If information is incomplete, obtain the missing information first rather than directly generating application recommendations.

Follow references/admission-estimation.md. Prioritize comparisons by rank. Output an interval probability only when there are at least three comparable years, the current year’s enrollment plan has been published, the definitions are consistent, and the sources are verifiable. Otherwise, output only qualitative levels—“Reach / Possible / Relatively Safe / Safety”—and clearly state which data are missing.

Never use aggregators, self-media accounts, forums, or cutoff scores without a stated year as key evidence. Critical admissions facts must be checked against the provincial education examination authority, Sunshine Gaokao, or the university’s official undergraduate admissions website.

If the user provides only a score and no rank, the Agent should first attempt to retrieve the corresponding year’s official score-to-rank table from the provincial education examination authority. If data for the current year have not been published, the most recent year may be used to estimate a “reference rank,” but it must be explicitly stated that this cannot replace the actual rank. Only when no official source can be found, the page is no longer accessible, or data for the user’s province are unavailable should the user be asked to upload or provide a score-to-rank table, enrollment plan, or cutoff-score file.

The final report should preferably be rendered with the script:

python3 <skill-dir>/scripts/render_report.py \
  --workspace <candidate workspace absolute path> \
  --output <candidate workspace absolute path>/final-report.html

The script reads , , , and from the workspace, and applies assets/final-report-template.html. When fields are missing, display “To be supplemented / Unverifiable”; never fabricate content.

The final report should first present the key conclusions: whether the direction is a good match, whether the current data are sufficient, and what to add next. Put the full personality scores, detailed major and industry information, methodology, uncertainties, and source ledger in expanded collapsible sections later, so that the user is not overwhelmed by technical terminology as soon as they open the report.

The final report must include at least the following:

If the user requests a review, the report will be used for an important decision, or the Agent is uncertain about a key source, launch a fresh subagent to conduct an independent audit according to AUDIT-FORMAT.md. If the audit does not pass, do not display “Approved for formal submission.”