Communitygithub.com

Fldicoahkiin/lab-report-autopilot

Automatically run Chinese university computing labs, capture results, and fill the supplied DOCX template across Windows, Linux, and macOS.

Was ist lab-report-autopilot?

lab-report-autopilot is a Claude Code agent skill that automatically run Chinese university computing labs, capture results, and fill the supplied DOCX template across Windows, Linux, and macOS.

Funktioniert mitClaude Code~Codex CLI~Cursor
npx skills add Fldicoahkiin/lab-report-autopilot

In Ihrer bevorzugten KI fragen

Öffnet einen neuen Chat, in dem dieser Agent-Skill bereits geladen ist.

Dokumentation

Lab Report Autopilot

Drive the report from assignment to delivery: inspect the files, choose the environment, run the experiment, capture the useful results, fill the user's template, and inspect the finished DOCX. Once the required files and personal fields are available, continue without asking the user to perform routine steps.

Never invent software output, screenshots, user details, citations, or environment values. Interrupt only when an unavailable environment, missing authorization, or missing required field would make the report false.

Set up

Resolve the directory containing this SKILL.md as SKILL_DIR. Run bundled scripts by absolute path; the current directory may be the user's report folder rather than the skill folder.

If no report workspace exists:

python3 "$SKILL_DIR/scripts/init_report.py" work/report-name \
  --kind lab \
  --discipline packet-analysis \
  --title "ARP 协议抓包与字段分析"

Use --kind practice for professional practice. A supplied DOCX remains the report template; do not replace it with the generic Markdown draft.

The image checker needs Pillow:

python3 -m pip install -r "$SKILL_DIR/requirements.txt"

Read the material

Classify each input before writing:

  • Template: controls sections, fields, page layout, and visual style.
  • Good reference: shows useful step length, wording, or screenshot placement.
  • Counterexample: shows problems to avoid.
  • Task material: supplies the required data, code, topology, and expected result.
  • Private material: follows the user's handling instructions and supplies only authorized facts. Read user-data-and-privacy.md.

Learn the pattern, not another person's identity or results. Do not copy names, screenshots, addresses, private records, or first-person claims.

Read disciplines.md, then load the one matching discipline file. Load a second only when the assignment actually crosses subjects.

Decide what must be delivered

Record:

  • required sections and grading prompts;
  • fields the user has supplied;
  • software, operating system, data, topology, and expected output;
  • the visible result needed for each submitted claim;
  • length, date, filename, and file-format rules.

Fill names, student numbers, class, teacher, organization, role, dates, and other fields only from values the user supplies or confirms. Use the same value everywhere it appears. Ask once for missing fields that the template requires. Leave signatures, scores, and approval fields empty unless the user explicitly provides them. Do not reuse a filename pattern from an unrelated report.

Before editing Word, read docx-qa.md and measure the supplied template: paper size, margins, tables, fonts, sizes, line spacing, indents, image widths, captions, headers, footers, and page numbering. These measured values control the output.

Choose the environment

Read environment-playbook.md, then inspect the current machine:

python3 "$SKILL_DIR/scripts/probe_environment.py" --json

Do not assume the author's computer or virtualization product. Follow the operating system and software named by the assignment. For common Chinese university coursework, treat Windows as the main route for Office, PowerShell, SQL Server, eNSP, vendor simulators, and Windows-only teaching software. Use Linux or Kali for Linux networking and security tools. Use macOS natively only when the named software and required screenshots remain equivalent; otherwise use any available Windows or Linux VM, container, WSL, or authorized remote lab.

Keep permission fixes and failed attempts in working notes unless they change the submitted result. Security work must stay inside local labs, course systems, deliberately vulnerable targets, or other authorized systems.

Run and capture

Read evidence-and-screenshots.md and the selected discipline file.

  1. Build the smallest setup that satisfies the task.
  2. Run every command and code path that will appear in the report.
  3. After a meaningful GUI action, inspect the fresh application state.
  4. Record the exact inputs, filenames, versions, addresses, frame numbers, and observed values.
  5. Capture the screen that shows the result; do not capture every click.
  6. Keep original screenshots, packet files, data, and source code unchanged.

Generated diagrams may explain a concept but may not stand in for a real software result. Screenshots used in one report must agree on filenames, addresses, ports, values, and environment.

Check the image folder before drafting:

python3 "$SKILL_DIR/scripts/evidence_manifest.py" path/to/evidence \
  --output evidence-manifest.json

Fix corrupt files, accidental duplicates, unreadable captures, exposed private data, and missing result screens.

Write from the observed result

Read writing-and-code.md.

For each practical step, state the operation, the value or behavior observed, the nearby screenshot, and a short explanation only when it helps. Keep theory and summaries proportional to the examples and the assignment.

For professional practice, write it as assigned work: what was received, what was checked or operated, how the judgment was made, and what result was recorded. Do not add presentation notes about screenshot counts, report structure, retained artifacts, or what the report chooses not to introduce. Do not add a company homepage merely to fill a source line.

For code, implement only the requested behavior. Do not add unused abstractions, speculative options, impossible-case checks, or comments that restate simple code.

Build and check the DOCX

Use the host agent's document tools when available; otherwise use compatible DOCX tooling and a real renderer. Fill the supplied template in place. Remove sample text only where the report content replaces it.

Render the latest DOCX and inspect every page at 100%:

  • screenshots and captions are readable and aligned;
  • text, tables, and images do not clip or overlap;
  • Chinese and Latin glyphs render correctly;
  • headings, numbering, spacing, indents, margins, headers, and footers match the template;
  • names, dates, commands, addresses, frame numbers, figures, and conclusions agree;
  • no extra blank page or nearly empty carried-over page remains;
  • the filename follows this assignment's rule;
  • no credentials, private workplace data, or unrelated personal information remain.

Repeat the render after every layout change. Deliver only the requested files unless the user asks for working material.

Verwandte Skills