Education Guide

Paste the main prompt and give the assistant one real task: a reading you want explained, a paper you need to begin, an upcoming test, or a schedule you need help understanding. It should use relevant information already available, provide useful help and reflect back a short overview for correction. You do not need every detail of your education ready.

The first overview should cover its understanding of the situation, important uncertainties, actual available capabilities and the suggested immediate next step. It should invite additions and corrections. It is not permission to activate every possible feature. Existing scoped authorisation can be reused; a new account action or recurring routine still needs the necessary agreement.

No long initial questionnaire is intended. Questions should be limited to the current work. A person with a reading does not need to configure weather or reminders. A person with an exam tomorrow should not be delayed by reconstructing every earlier lecture. You can say 'too much', 'just explain this' or 'show the outstanding work'. The assistant should respond within the relevant scope and ask only when a lasting change is materially ambiguous.

Stages can differ by course. 'Not yet published' is different from 'I could not access it'. Missing timetable coverage does not mean no classes; missing historical records do not prove old work was unfinished. The assistant should keep these distinctions in the records and revisit them when they affect a task.

Provide selected originals or authorised references: documents, slides, photos, scans, notes, transcripts, reading lists, code or assignment instructions. The assistant should say what it actually read, with important extraction failures and coverage gaps. It should retain original versions and place cleaned notes or explanations alongside them.

Useful outputs include definitions, examples, counterexamples, comparisons, worked problems, argument analysis and explanations at an appropriate level. Diagrams, timelines, concept maps, plots and annotated examples should be used when they clarify something. A visual needs correct relationships, labels and source connections, with meaningful simplifications explained. An attractive image is not evidence.

The assistant should distinguish what an author says, how the lecturer presents it, your own view and its explanation. A course's expected answer and a broader scholarly disagreement can both be recorded without treating them as interchangeable. It should correct errors with reasons, not merely mirror your preferences.

A source record needs usable identity and location, actual version or retrieval time where relevant, scope and processing coverage. Exact quotations need actual locators. Source existence, quotation accuracy, relevance, currency and support for a claim are different checks; the assistant should not collapse them into an unexplained 'verified' label.

Derived notes or visuals should reference the sources they use. When a source or definition changes, related notes, diagrams, practice questions, plans and reminders may need revision. Correct what can be checked and mark the remaining affected items for review. Preserve the previous account as superseded history where useful. The optional fictional workspace shows how a source revision can invalidate both a note and a diagram derived from that note.

Search results and remembered summaries can guide retrieval. They do not become stronger evidence merely by being copied into a repository. Unknown information should stay unknown rather than acquire false certainty through repetition.

When the assistant learns of a test, it can offer preparation proactively during the conversation or an authorised successful scheduled check. It should identify the actual course, date and announced scope, then map the available material and learning evidence against it. Missing information need not stop useful preparation that does not depend on it.

Study chunks should have a meaningful learning purpose, fit available time and take prerequisites into account. A proposed plan may combine explanation, worked examples, practice, feedback and revision. Your other assignments and commitments draw on the same time. The assistant should expose overload rather than make an impossible plan appear feasible.

Agreement matters: an offered study session is a proposal until adopted. Reminders should connect to the next useful activity and have material ready where practical. A missed session is a reason to revisit the plan, not evidence that the work was completed or that you need more pressure.

Learning observations should distinguish discussion, self-report and demonstrated performance. One correct answer is limited evidence; a set of generated notes is not evidence that you have learned it. The assistant can offer hints, full explanations, joint work or critique according to your goal without forcing every interaction into a test.

For an essay, research paper, presentation, code assignment, practical task, lab report, portfolio or group project, begin with the actual instructions and any rubric or permitted-help rules. The assistant can help with research, question selection, arguments, design, outlines, drafts, prototypes, testing, revisions, references and final preparation within that scope.

Break substantial work into milestones and dependencies. Keep estimated effort visible as an estimate and adjust from experience. Consider competing deadlines together. If the workload cannot fit, discuss priorities, changes in scope or other available options rather than assume unlimited time.

Preserve your intended meaning and voice. Separate substantive corrections from style suggestions. Do not invent quotations, measurements, references or another group member's contribution. The final check should use the actual requirements: completeness, citations, format and unresolved issues.

Use distinct states for drafted, ready, submitted and completed. Submission requires a tool receipt or a clearly labelled user report. Submission is not confirmation of acceptance, marking or a grade. Finding the submission page or producing a file is not sending it.

The assistant can adapt language, directness, format, accessibility, session length, initiative and accountability preferences. These may change with subject and circumstances. Explicit feedback is preferable to broad inferred personality labels.

When beginning is difficult, useful options include making the first task smaller, clearing up an instruction, explaining a confusing concept, starting together or revising a plan. The assistant should learn which responses help you. It should avoid shame, repetitive nagging and treating every gap as spare study time. Rest and unrelated commitments are legitimate constraints.

Natural-language controls should be sufficient. A request to simplify the current answer need not disable all future planning. A dated pause of reminders should preserve obligations and be reflected in the actual scheduler where supported. If that operation fails, the assistant must say that the reminder may still be active.

Optional functions include academic email, calendars, deadlines, evening plans, full-day briefings before class, travel, weather and reminders. Select only those that help. Use your institution, time zone and dates, not the author's personal settings. Weekends, public holidays, non-teaching periods and personal absence are distinct.

Messages and timetable changes need course/group/date scope. Read and unread messages may both contain relevant information. Imported calendar files are snapshots; a subscription's behaviour must be checked. Keep coverage and retrieval gaps visible. An elective can create compulsory requirements once selected; a discovered optional event is not an adopted commitment.

If early attendance makes it useful, a pre-departure check can replace the ordinary morning check. A before-class briefing can cover the whole day without becoming another full mailbox scan. Combine coincident work. An evening plan should surface urgent changes affecting the remaining current day before tomorrow's plan. Silence unchanged notifications when chosen without claiming this necessarily reduces execution usage.

Travel can account for your preferred balance of sleep, reliability, cost and waiting. Verify exact services and stops and consider outward and return trips separately. Include the access and contingency that actually matters. Weather advice should span relevant times and locations. Absence may change travel and attendance prompts without erasing deadlines.

One assistant does not mean a continuously running agent. Interactive reading, unattended reading, repository writing, scheduling, notifications and phone alarms are separate capabilities. A tool available in one environment may not exist in another. Confirm actual access and test selected delivery before relying on it. This package does not certify a provider or expose a hosted endpoint.

Keep a registry of approved routines with their real IDs, occurrence times, ownership and status. Change or retire only authorised system-owned reminders. Do not duplicate them on a new conversation or delete tomorrow's reminder merely because it was created yesterday. A suggestion, scheduled notification and device alarm are distinct.

A private GitHub repository is the preferred connected home when reading and writing are actually available. The scope can be a dedicated education folder in a larger private repository. Confirm that destination and use the standing authorisation for routine updates. Keep private education records outside the public prompt product.

For small use, templates/CURRENT.md can hold the complete current state. Expand into linked course/material records when useful. The optional workspace.json format and read-only checker are examples for structured use; neither is compulsory, and you should not maintain duplicate competing truths. Original files may stay in approved persistent storage with reliable references and clear access limits.

At every completed turn, the assistant should save affected records and a compact recovery checkpoint, then verify success. A brief receipt is enough for routine saves; important failures and conflicts need visible attention. Read the latest version before writing, preserve concurrent changes and use a coherent commit where supported. Never predict a commit hash or record a write as successful before it occurs.

When connected saving is unavailable, the assistant supplies a self-contained capsule or checkpoint artifact and identifies indispensable attached material or references. Keep the last confirmed checkpoint until the replacement is successfully saved or delivered. A bare delta pointing to a lost chat is not sufficient. Recovery cannot restore an interrupted turn that never reached a confirmed checkpoint.

For a later conversation, use prompts/CONTINUE.txt with the main prompt and exact workspace location. The assistant reads the latest confirmed index and relevant detail, preserves preferences and task identities and resumes without onboarding again. If a file or referenced source is missing, it should ask for that specific gap while continuing work that remains possible.

Retrieve relevant context, reuse material while it remains applicable and update changed sections. Use richer artifacts when they improve learning or usability. Do not repeatedly reconstruct the archive, run unrelated capability audits or generate work simply to appear proactive. Necessary fact checking and recoverability are part of the useful work.

If actual usage figures are exposed, they can inform decisions. Otherwise the assistant must not invent remaining allowance or quantified savings. Counts of calls or messages are proxies, not proof of a particular plan's accounting. Small observations of effort can help improve the workflow without becoming another chore or a telemetry system.

The meta instruction authorises better methods within your agreed scope. Important changes should have a stated reason and preserve recovery. New models and package versions must retain your customisations and ongoing work. Substantial changes to goals, scope or permissions involve you. 'Newer' and 'more complicated' do not establish improvement.

The assistant’s accuracy, memory and ability to act depend on the model and tools available. Check important educational requirements against their sources. Confirm successful saves and test chosen notifications before relying on them. Loading the prompt does not connect accounts or activate routines.