Thesis Review Prompt Pack

Use With Codex / Claude Code / Antigravity

This is the recommended path because a thesis review needs file access, missing-input tracking, reference evidence, and persistent review notes. The goal is student review and feedback, not automatic rewriting of the source work.

Back to Home ไทย

Install and Start

1. Choose a workspace AI tool

Use Codex CLI, Claude Code, Antigravity, or another AI tool that can read and write files in a local workspace.

2. Create a review workspace

Run one of the setup command blocks below. The default project folder is my_thesis_review/.

Ground Rules

Create review notes first. Do not edit the thesis source immediately.

Do not let AI invent citations, DOI, venue rankings, metrics, results, IRB/legal requirements, or legal statements.

The reviewer should point out weaknesses, recommend next checks, and then let the author do the revision work. If students use AI again, they should practice writing their own prompts from the review notes.

Why this saves tokens

Do not ask the AI to read the whole LaTeX folder or every file at once. Put mixed files in 00_inbox/, let the AI organize them, then use 00_review_instructions/ as the selected context for later review rounds.

Step 1: Create the Workspace

Linux / macOS

wget -O install-thesis-review.sh https://punpiti.github.io/thesis-review-prompt-pack/install-thesis-review.sh
bash install-thesis-review.sh my_thesis_review

Windows PowerShell

Invoke-WebRequest -Uri "https://punpiti.github.io/thesis-review-prompt-pack/install-thesis-review.ps1" -OutFile ".\install-thesis-review.ps1"
PowerShell -ExecutionPolicy Bypass -File ".\install-thesis-review.ps1" -Project "my_thesis_review"

These scripts only download the prompt pack, extract it, and create the workspace from the template. Students still place their own files and use review notes to think through the work.

Put all mixed input files in my_thesis_review/00_inbox/: thesis drafts, LaTeX, Word, PDF, BibTeX, reference PDFs, figures, tables, results, scripts, logs, metrics, and baselines.

Step 2: Ask AI to Organize the Inbox

init workspace จากไฟล์รวม

This Thai trigger is intentionally short. Workspace instructions in AGENTS.md and CLAUDE.md define what it means.

Full English version

Initialize this thesis review workspace from the mixed files in 00_inbox/.

Create missing folders if needed. Copy files from 00_inbox/ to the appropriate target folders, but do not delete the original inbox files.

Use this classification:
- latest thesis/manuscript drafts -> input/v02_revised/
- older/original drafts -> input/v01_original/
- readable .tex/.md/.txt thesis text -> 01_working_text/
- .bib/.ris/.enw/reference lists -> 02_references/bib/
- reference PDFs -> 02_references/pdf/
- DOI pages, publisher pages, official abstracts -> 02_references/official_pages/
- datasets, tables, figures, logs, scripts, metrics, baselines, result evidence -> 03_results/

If classification is uncertain, leave the file in 00_inbox/ and mark it as needs classification.
Create or update 00_review_instructions/file_inventory.md with source path, target path, file role, and uncertainty.
Do not edit thesis/manuscript/reference content in this step.

Step 3: Build Review Instructions

เตรียม instruction จากงานปัจจุบัน

Full English version

Prepare review instructions from the current work.

Load 00_review_instructions/.
Read the latest thesis/manuscript files from input/ and 01_working_text/.
If the work is LaTeX, identify the current root .tex file and follow \input{} and \include{} only as needed. Identify bibliography, figure, table, and result files referenced by the manuscript.

Create or update:
- 00_review_instructions/review_profile.md
- 00_review_instructions/document_map.md
- 00_review_instructions/review_scope.md
- 00_review_instructions/missing_inputs.md

Use workspace evidence only. Mark unavailable facts as missing input.
Do not invent citations, DOI, venue rankings, results, IRB/legal requirements, or claims.
Do not edit thesis/manuscript/reference files.

Step 4: Run the Review

ประเมิน

This short trigger asks the workspace AI to continue from prior review state if available, create review notes, and avoid editing the source thesis.

Full English version

Review this workspace as a full thesis/manuscript review, but do not edit the source thesis yet.

Read available context from 00_review_instructions/, 01_working_text/, 02_references/, 03_results/, prompt_pack/, and 04_review_notes/99_review_state.md if it exists.

Create or update:
- 04_review_notes/00_inventory.md
- 04_review_notes/01_full_review.md
- 04_review_notes/02_reference_audit.md
- 04_review_notes/03_student_action_checklist.md
- 04_review_notes/99_review_state.md

Use prompt_pack/01_full_thesis_review_prompt.md and prompt_pack/02_reference_bibtex_prompt.md as the review criteria.
Use templates from prompt_pack/templates/04_review_notes/ when helpful.

Check research logic before language polishing: problem survival, genuine failure, mandatory gates, measurement ceiling, and no-new-problem guardrails.
Separate output from outcome, and trace the argument chain: problem -> prior attempts/evidence -> unresolved need -> objective/RQ -> alternatives/selection rationale -> method -> result/evidence -> output -> outcome/by-product -> conclusion.
Do not invent unsupported facts. If evidence is missing, write missing input.
Record what was reviewed, what remains unfinished, and the next prompt to continue.

Step 5: Deep Review

ประเมินเชิงลึก

Advanced stage only. New students should run the normal review first, fix blocking issues, and use this later when the draft has enough references, results, and review notes to stress-test.