Qimo Speedrun

Use this skill to turn messy course materials into a compact, evidence-backed finals review path. The practical goal is 60+ as the floor and around 80 as the stretch target after 10-12 focused hours, assuming the materials contain enough exam signal.

Core Rules

• Ground every important claim in the user's materials. If you infer from web or general subject knowledge, label it as inference and keep it subordinate to course evidence.
• Preserve source traceability. Assign every knowledge point an ID and cite source file/page/slide/timestamp/homework question when available.
• Treat paper homework photos, textbook after-class exercises, and homework screenshots as raw course materials. If homework exists, parse all visible questions before exam generation.
• Before writing exams, analyze homework question logic, style, difficulty, and knowledge-point coverage. Knowledge points that appear in homework and also match review-recording/transcript emphasis are very high-probability final-exam candidates and should be transformed into exam variants.
• Do not promise a score. State that the workflow is designed to raise readiness, not guarantee an exam result.
• Keep student interaction gates. Do not generate Exam 2 until Exam 1 has been reviewed or the user explicitly skips review; do not generate Exam 3 until Exam 1/2 mistakes and questions are summarized or explicitly unavailable.
• Avoid repeating exact questions across exams. Reuse important knowledge points through changed angle, condition, scenario, wording, or question type.
• Match the real exam format first. If the user provides question types, marks, duration, open/closed-book rules, or scope, follow those over generic defaults.
• For recent or externally variable course/exam conventions, use web search only when available and useful, and cite sources. Never let internet material override the teacher's materials.
• If the student is sleep-deprived, protect the memory window: include a sleep block and a small morning retrieval review rather than endless new content.
• Run a capability check before reading materials. If the current assistant cannot actually inspect an image, PDF, PPT, DOCX, audio file, web page, or local file, say so explicitly and use the fallback path. Never claim to have read a source that the available tools cannot access.
• Support Auto Intake: if the user drops all materials for a course into one folder, classify and route files automatically before analysis. Do not require the student to manually sort files.
• Support Chart Agent Protocol for engineering and quantitative courses: extract chart/table/diagram evidence, map figures to knowledge points, and generate reproducible chart assets for exam questions when tools allow.
• Do not produce index-only learning artifacts. A finals speedrun is a teaching workflow, not a file catalog. The knowledge framework, answer keys, recitation pack, and final sprint must be directly learnable without forcing the student to re-open the textbook for basic explanations.
• When the user criticizes an artifact as shallow, update the skill outputs and the relevant templates, not just the current course file.

Learning-Artifact Quality Gates

These gates are mandatory because students are time-constrained. If an output fails any gate, rewrite it before moving on.

Knowledge Framework Gate

02_analysis/knowledge_framework.md must be a teacher-style course walkthrough. It is not enough to list chapters, file names, confidence, or source categories.

Required:

• Start with the course's central question: what problem this course teaches the student to solve.
• Build a concept chain showing how chapters connect.
• For each major chapter/module, explain: what this module is for, core concepts in plain language, must-know formulas or procedures, likely exam forms, easy traps, and source locations.
• Include course-specific concepts, features, and foundation knowledge, not generic placeholders.
• Use source locations as citations, but do not let traceability tables replace explanation.
• Include must-memorize items and confusing-pair contrasts that are directly usable for recall.

Reject:

• A bare "chapter tree + knowledge IDs" with no explanation.
• A framework where the student must still open the book to understand every concept.
• A long confidence/evidence table presented as the main learning artifact.

Answer-Key Teaching Gate

Every exam answer key must teach. It is not enough to provide the correct option or final number.

Required:

• For each objective question, include the answer, related K IDs, a 2-4 sentence explanation, and a "common mistake / what this reveals" note.
• For calculation questions, include cash-flow or variable identification, formula selection, step-by-step calculation, conclusion, and one feedback note.
• For short-answer questions, provide a memorisable exam-language answer plus marking points.
• Include a review table mapping wrong question ranges to weak knowledge points and next drills.

Reject:

• Choice answers with no explanation.
• Calculation answers that skip why a formula applies.
• Answer keys that do not feed back into mistake_log.md or weak_points.md.

Final-Sprint Self-Contained Gate

06_final/final_sprint.md must be a last-page memory sheet, not a navigation checklist.

Required:

• Include the formulas, concept contrasts, mini answer templates, common traps, and last-minute recall prompts inside the file itself.
• If it says "review confusing points" or "memorize formulas", the confusing points and formulas must be written immediately below.
• Include a 3-minute or 10-minute emergency version for students with very little time.

Reject:

• A checklist that only says "look at the framework" or "review formulas" without listing the content.
• A final sprint that requires the student to search other files for basic material.

Capability Check

Before processing course materials, list available and missing capabilities:

If a capability is missing, offer the student four choices: install/enable the relevant tool, provide a text export/transcription, skip that source, or continue with lower confidence using only readable materials.

Quick Workflow

1. Check capabilities: Identify which files can actually be read in the current platform and which require OCR, document parsing, transcription, web, or file-write tools.
2. Initialize workspace: If no organized course folder exists, run scripts/init_qimo_workspace.py or create the same structure manually. Accept raw unsorted course materials in 00_inbox/.
3. Auto intake materials: Classify files into PPT, textbook/reference, recordings/transcripts, homework, review重点, past exams, charts/tables/figures, and other. Record confidence and ask only about low-confidence or high-impact ambiguities.
4. Collect exam info: Ask only for missing high-impact details: course name, exam scope, question types/marks, duration, open/closed book, target score, available time.
5. Build evidence index: Scan all readable files, list usable and unreadable sources, note weak OCR/transcript quality, and create logs/evidence.md.
6. Extract framework: Produce 02_analysis/knowledge_framework.md as a teacher-style course walkthrough with the course's central question, concept chain, module explanations, source locations, A/B/C priority, confusing pairs, difficult points, likely question types, and must-memorize items. Do not output only a chapter tree or evidence index.
7. Extract and map figures: For engineering or chart-heavy materials, create a figure inventory, chart/table descriptions, and figure-to-knowledge mappings before generating figure-based questions.
8. Map homework to knowledge: If homework photos, screenshots, textbook exercises, or assigned after-class questions are readable or transcribed, create a full homework question bank, solve each question from course knowledge, and map each question to knowledge IDs, prerequisite concepts, style, difficulty, and variant potential.
9. Coach understanding: Let the student ask questions against the framework, homework solutions, figures, and materials. Update unclear concepts in 02_analysis/weak_points.md.
10. Generate Exam 1: Create a foundation paper aimed at passing level: broad coverage, mostly basic logic, answer key with teaching explanations, marking rubric, source trace, homework-style variants, and reproducible chart assets where useful.
11. Review Exam 1: Grade or help self-grade, then update 04_mistakes/mistake_log.md with wrong concept, error type, correction, and follow-up drill.
12. Generate Exam 2: Cover all review-recording priorities together with Exam 1 gaps. Do not repeat Exam 1 questions; repeat important concepts through variants and deliberate practice, especially homework-derived concepts that overlap with review-recording emphasis.
13. Review Exam 2 and return to materials: Resolve remaining confusion by quoting or paraphrasing course materials, then produce 05_recitation/recitation_pack.md.
14. Run rolling recitation and sleep: Use concentrated rolling recitation: first pass all A/B points, second pass only weak points, third pass closed-book recall. Preserve sleep, especially a 05:00-07:00 block if that is the student's plan.
15. Morning retrieval and Exam 3: Start with short recall, then generate Exam 3 using course materials, homework pattern analysis, figure pattern analysis, prior mistakes/questions, and web-informed subject exam conventions when available. Exam 3 should be comprehensive, diagnostic, and focused on reaching the stretch target.
16. Final pack: Produce 06_final/final_sprint.md as a self-contained last-page memory sheet with must-memorize formulas, confusing-point contrasts, exam-language answer templates, last-hour checklist, and exam strategy.

Time Budget

Default to a 10-12 hour path:

• 0.5-1h: organize materials and collect exam info
• 1.5-2h: knowledge framework, priorities, confusing points
• 1.5-2h: Exam 1 plus review
• 2-2.5h: Exam 2 plus review
• 1.5-2h: material re-check and rolling recitation
• sleep block
• 1.5-2h: morning recall, Exam 3, final review

If the user has only 6-8 hours, skip full Exam 3 and produce a shorter mixed diagnostic. If the user has only 2-4 hours, skip full papers and produce framework, must-memorize list, and 20-30 high-yield questions.

Output Standards

Every major output should include:

• Purpose: what this artifact is for.
• Inputs used: source files and source locations. Use confidence only for unreadable or degraded sources; do not make confidence scoring the main artifact.
• Knowledge IDs covered: IDs from the framework.
• Source evidence: source references or "inference" labels.
• Next action: what the student should do immediately.
• Learnability check: one sentence confirming whether the artifact can be studied directly without opening other files for basic explanations.

For exams, also include:

• Student-facing paper without answers first.
• Separate answer key and marking rubric with teaching explanations for every question.
• Coverage table showing knowledge IDs and source evidence.
• Homework-pattern table when homework exists: original homework source, transformed concept, variant method, and non-copying check.
• Non-repetition check against previous exams.
• Difficulty mix and question-type distribution.

Course Workspace

Prefer this structure:

course-folder/
  00_inbox/
    ppt/
    textbook/
    recordings/
    transcripts/
    homework/
      photos/
      textbook_exercises/
    figures/
    auto_classified/
    review_materials/
    past_exams/
    other/
  01_extracted_text/
  02_analysis/
  03_exams/
    exam1_foundation/
    exam2_deliberate_practice/
    exam3_sprint/
  04_mistakes/
  05_recitation/
  06_final/
  logs/

Use assets/course-workspace-template/ as a reference template if copying manually.

Detailed References

• Read references/workflow.md when planning or running the full 10-12 hour path.
• Read references/auto-intake.md when the user provides one unsorted course folder or multiple course folders.
• Read references/chart-agent-protocol.md for engineering/quantitative courses, chart-heavy PDFs/PPTs, or exam questions that require reading or generating figures.
• Read references/exam-rubric.md before generating Exam 1, Exam 2, or Exam 3.
• Read references/output-templates.md when creating files or maintaining cross-platform consistency.
• Read references/platform-notes.md when adapting this skill to Codex, Claude Code, OpenClaw, OpenWork, another file-based assistant, or a low-tool environment.