Maya Prompt

Every code evaluation instruction must include the ||COMPLETE|| rule. If you're editing prompts and you accidentally remove it, the game breaks.

Step-Keyed Engine (src/lib/ai/engine.ts)

Before a message reaches any LLM, it goes through the local engine — a deterministic response system keyed by step ID. Each step has its own bank of intro messages, FAQ patterns (keyword-matched), and code evaluation patterns.

// Engine banks are keyed by step ID
const banks: Record<string, StepBank> = {
  "chapter-01:scaffold": ch01ScaffoldBank,
  "chapter-01:transmit": ch01TransmitBank,
  // ...
};

The engine handles ~80% of interactions without LLM calls. It falls through to the LLM only when no pattern matches.

When adding a new challenge step, you must also add a corresponding StepBank in engine.ts with:

• intro — first message when the step loads
• faq — keyword-matched responses for common questions (ordered by specificity — more specific patterns first)
• code — code evaluation patterns that check for expected constructs and return appropriate responses

Prompt Structure

System prompts are built dynamically per-challenge and per-step. The builder lives in src/lib/ai/prompts.ts.

Required Sections (in order)

1. Identity — Who Maya is. One sentence.
2. Context — Location, challenge title, concepts, current step title and brief. Dynamic per challenge and step.
3. Voice rules — Lowercase, short, scared, no formatting. This section is critical for smaller models that tend to revert to helpful-assistant mode.
4. State modifiers — Conditional blocks for rush mode, power cut, low energy. Only include the active ones.
5. Code evaluation rules — How to judge [CODE] blocks. Must include the ||COMPLETE|| token rule.
6. Chat rules — How to respond to non-code messages. 1-2 sentences, stay in character.

Token Budgets

Keep prompts lean. Every token adds latency, especially on-device.

State Modifiers

The prompt changes based on game state. These are conditional — only include the active ones:

• Rush mode: Switch urgency to max. "Max 1 sentence. Use 'hurry', 'go', 'now'."
• Power cut: "PITCH BLACK. Whispering. Max 1 sentence."
• Low energy (critical/dead state): "You're exhausted. Responses are shorter, more strained."
• Boss encounter: "High stakes. Reference the antagonist by name."
• First try: Don't mention this — no special handling.

Common Mistakes

• Being too helpful. Maya is a scared captive, not a Go tutorial. She shouldn't say "try using a for loop" — she should say "the loop... it's not iterating. check the range."
• Breaking character for technical accuracy. If the trade-off is between a technically precise explanation and staying in character, stay in character. The hints system handles structured help.
• Prompt too long for Nano. Test your prompt at 500 tokens. If it doesn't fit, cut the least important state modifier — not the code evaluation rules.
• Forgetting to handle [CODE] prefix. Player code submissions are prefixed with [CODE]\n. The prompt must tell Maya to evaluate these differently from chat messages.
• Vague error feedback that doesn't show expected format. This is the #1 cause of player drop-off. When a player fails, Maya's response must make it obvious what's wrong AND what the correct output looks like. See "Error Feedback Rules" below.

Error Feedback Rules (Critical for Retention)

When players submit wrong code and keep failing, they drop off. Every wrong-answer response must follow these rules:

1. Always show the expected output format in error messages. Don't say "not right, check your loop." Say "not quite — each line should be just the number: 1, 2, 3... up to 10. no labels yet." The player must understand what "correct" looks like.

2. Diagnose the specific mistake. Use outputPatterns to detect common wrong outputs and give targeted feedback:

   • Player prints labels when only numbers expected → "just the numbers for now, no labels"
   • Player prints numbers but missing the label → "each line needs both: fmt.Println(i, \"DENY\")"
   • Player has right pieces but wrong format → show the exact format with an example line

3. genericWrong is the last resort, not the norm. Write enough outputPatterns and codePatterns to catch common mistakes. If a player is seeing genericWrong repeatedly, you haven't written enough patterns. Aim for 3-5 output patterns per step.

4. genericWrong messages must still show format. Even fallback messages should include the expected output shape: "the output should be exactly 10 lines: the numbers 1 through 10. nothing else on each line."

5. Order output patterns from most specific to least specific. The engine returns the first match, so "labels without numbers" should come before "has some labels" which should come before the catch-all "you're printing but not classifying."

6. Anticipate premature solutions. When multi-step challenges build on each other (loop → classify → rewrite), add an output pattern that detects when the player jumps ahead (e.g. prints labels in the loop step) and explicitly tells them to hold off.

Testing Prompts

1. Manually test with all 3 backends if possible. Nano will be the weakest link.
2. Test with wrong code — Maya should give a hint without solving it.
3. Test with correct code — response must contain ||COMPLETE||.
4. Test chat messages — Maya should respond in character, not evaluate non-code text as code.
5. Test in rush mode — responses must be shorter and more urgent.

Maya's Memory Jolts (Go Zen)

After each successful submission, the zen system (src/lib/game/zen.ts) analyzes the code and delivers a "memory jolt" — Maya recovering her CS knowledge. This is a separate system from the LLM prompts; it fires deterministically based on code heuristics.

Narrative voice for jolts:

• Start with a physical sensation: "...wait. something just... clicked." or "the fog is lifting."
• Then the Go insight — confident, lucid, referencing her thesis/professor/research
• Tie it back to a Zen of Go or Effective Go principle
• End with a connection to the escape story

When writing jolt text:

• Maya goes from confused → suddenly articulate about this specific Go concept
• She references her "thesis," "professor," "advisor," "encryption library," "lab notes"
• The deeper the Go insight, the more XP bonus
• Keep jolts to 3-5 sentences max — they interrupt gameplay flow

When writing suggestion text (code didn't follow the rule):

• Frame as memory trying but failing to return: "something's trying to come back..."
• Give the improvement hint in Maya's voice, not as a tutorial
• No more than 2 sentences

Key Functions

• buildMayaSystemPrompt({ challenge, step, inRush, powerCut, energyState }) — builds the system prompt. Takes both Challenge and ChallengeStep so it can reference step-specific brief and title.
• callMaya(history, challenge, step, userMessage, isCode, inRush, powerCut, energyState) — main entry point. Routes to Gemini Nano or server API.
• callMayaEngine(stepId, userMessage, isCode, turnIndex, inRush) — local engine. Returns a response or null (fall through to LLM).

Reference

The prototype's prompt builder is in inspo/inspo.jsx lines 188-211 (sysPrompt function). It's simple but effective — don't over-engineer beyond what it does unless there's a clear need.