Thuta Learning
ExercisesAIbeginner

Context Engineering Challenge Drills

Relax. We'll talk through this in plain words — no textbook voice.

What you'll walk away with

  • Practice the Context Engineering Challenge Drills hands-on
  • Reinforce the skills you've already learned through practice
  • Learn to spot mistakes, fix them, and check your own work

Think about it this way for a second

This lesson steps up from the earlier drills — each task blends multiple concepts at once, drawing on chapters like Budget the Page, Summarize Don't Dump, Stale Notes Lie, and Missing Context and Ask Back, rather than testing just one skill. You'll face real trade-offs: with a token-limited context window, you have to decide which pieces of a large pile of information get priority. Other tasks train your judgment for catching stale data and for asking back instead of assuming when context isn't enough. At this level there's no single "best" answer anymore — how you balance the trade-offs is what matters most.

Exercises

Task 1 (Budget) - With a token budget of only 500 for the system, decide how much to trim from each of the 4 documents (word counts given) in the documents list, and what to include. Task 2 (Summarize) - Summarize the chat_history_sample into a single paragraph (3-4 sentences), keeping only the key decisions. Task 3 (Stale) - Compare the 3 notes in note_with_dates by date, find which one is stale, and explain why it shouldn't be trusted. Task 4 (Missing + ask back) - Identify the key fact missing from context for user_prompt_missing_context (which plan to upgrade to), and write one clarifying question instead of assuming.

Code example

json
{
  "token_budget": 500,
  "documents": [
    {"name": "Product spec v3", "words": 420, "updated": "2026-08-20"},
    {"name": "Product spec v1 (old)", "words": 380, "updated": "2026-05-02"},
    {"name": "Support FAQ", "words": 300, "updated": "2026-07-15"},
    {"name": "Team meeting notes", "words": 250, "updated": "2026-08-22"}
  ],
  "chat_history_sample": [
    "user: dark mode ထည့်ချင်တယ်",
    "assistant: OK, color scheme ဘယ်လိုလုပ်မလဲ",
    "user: system default နဲ့ toggle button ၂ခုလုံးလိုချင်",
    "... (message 8 ခု ကျန်, mostly UI detail ငြင်းခုံမှု)",
    "user: deadline ကို Sept 15 အထိ တိုးလိုက်ပြီ"
  ],
  "note_with_dates": [
    {"note": "Server region: Singapore", "date": "2025-11-01"},
    {"note": "Pricing: $9/month", "date": "2026-08-19"},
    {"note": "Server region: Frankfurt (migrated)", "date": "2026-08-05"}
  ],
  "user_prompt_missing_context": "ကျွန်တော့် account ကို upgrade လုပ်ပေးပါ"
}
You should see
Only relevant, up-to-date information should remain within budget, and a clarifying question should come out for the stale note and missing fact instead of an assumption.

5-minute try-it

Take just 5 minutes and do Task 3 — write down which note in note_with_dates is stale, and decide which one should actually go into context.

One quick word of caution

Letting the model guess at missing context is the riskiest move of all — asking back beats assuming and taking the wrong action.

Easy traps

  • Dumping in every document because "it's all important," blowing past the token budget
  • When two notes cover the same topic, grabbing whichever one appears first in the list instead of comparing their dates

Now try it yourself

Take just 5 minutes and do Task 3 — write down which note in note_with_dates is stale, and decide which one should actually go into context.

You'll know it worked when: Only relevant, up-to-date information should remain within budget, and a clarifying question should come out for the stale note and missing fact instead of an assumption.