Reading mode

Reading mode is a display preference. It uses a system font and gives words and lines more room.

Skip to the guide

Architect's Field Kit 04

Prompt Patterns

Reusable briefs for directing AI without handing over the thinking

Audience
Builders ages 13-18 and parents
Format
16 minute field read
Use
Keep open beside the builder's AI tool.
TCCA prompt frameworkTask, Context, Constraints, and Action form four connected quadrants around a clear direction.TTASKCCONTEXTCCONSTRAINTSAACTION
Prompt Patterns - visual field map

Field panel 01

compare

Extraction versus direction

The same AI can shrink thinking or strengthen it. The difference is the job the builder gives it.

  1. Extraction

    Give me the finished answer

    The goal is hidden, judgment is skipped, and the builder cannot defend the result.

    Evidence: Looks finished, but the thinking trail is empty.

  2. Direction

    Help me think and build

    The builder supplies the goal, material, constraints, checks, and final decision.

    Evidence: The trail shows questions, choices, tests, and revisions.

Field panel 02

framework

TCCA - the base brief

TCCA turns a wish into a usable direction. It works across AI tools.

  1. T

    Task

    State the exact job: compare, question, classify, debug, plan, transform, or critique.

  2. C

    Context

    Give the audience, situation, source material, prior attempts, and what success means here.

  3. C

    Constraints

    Set limits: format, length, tone, allowed sources, privacy line, and what the AI must not do.

  4. A

    Action and format

    Say what happens next and how the output should arrive so it fits the build.

Copy, adapt, direct

Task: help me compare two ways to solve [problem]. Context: the user is [person] and needs [outcome]. Constraints: ask before assuming; use only the material below; flag uncertainty. Action: give a tradeoff table, then ask me which route I choose.

Field panel 03

pattern

Pattern 01 - Clarify before acting

Use when the goal is real but the brief is still foggy.

  1. Use it for

    Missing requirements

    A real person asked for something, but important details are unclear.

  2. Direction

    Questions one at a time

    Ask the AI to identify what it needs to know, explain why each answer matters, and wait after each question.

  3. Check

    End with a brief

    The AI summarises the agreed goal. The builder corrects it before any making starts.

Copy, adapt, direct

You are helping me turn a rough idea into a build brief. Ask one clarifying question at a time. Do not design the solution yet. When the goal, user, constraints, and finish line are clear, summarise the brief for me to approve.

Field panel 04

pattern

Pattern 02 - Options with tradeoffs

Use when several routes could work and the builder needs to make a reasoned choice.

  1. Use it for

    Choosing scope or tools

    There are too many plausible directions and no perfect answer.

  2. Direction

    Request distinct routes

    Ask for three options that differ in speed, difficulty, risk, or usefulness.

  3. Check

    Choose in your words

    State the winning route and the tradeoff you accept. Do not let the AI choose for you.

Copy, adapt, direct

Give me three genuinely different routes to [goal]. For each, show the fastest useful version, main risk, skill stretch, and what evidence would tell me it worked. Do not recommend one until I state my priority.

Field panel 05

pattern

Pattern 03 - Stress test

Use before a public ship, important decision, or claim that needs to survive scrutiny.

  1. Use it for

    Finding weak spots early

    A draft feels good, but it has not met a hostile input or skeptical reader.

  2. Direction

    Attack the work, not the person

    Ask for likely failures, missing assumptions, counterexamples, and questions a skeptical user would ask.

  3. Check

    Verify every criticism

    AI criticism can also be wrong. Reproduce the issue before changing the build.

Copy, adapt, direct

Stress-test this against the goal and constraints below. Find five concrete failure cases. For each, give a way I can reproduce the problem. Do not rewrite the work. Wait for my test results before suggesting repairs.

Field panel 06

pattern

Pattern 04 - Debug one step

Use when something broke and random changes would make the trail harder to follow.

  1. Use it for

    A reproducible failure

    The builder can say what they expected, what happened, and how to trigger it again.

  2. Direction

    One hypothesis at a time

    Ask the AI to rank likely causes, then test the smallest distinguishing check first.

  3. Check

    Record cause, not just fix

    After the repair, rerun the original failure and a nearby case.

Copy, adapt, direct

Help me debug without guessing. Expected: [x]. Actual: [y]. Reproduce: [steps]. Give the top three hypotheses and the single smallest test that separates them. Stop after that test until I report the result.

Field panel 07

pattern

Pattern 05 - Check claims and evidence

Use for facts, numbers, quotations, links, or confident statements that affect the result.

  1. Use it for

    Evidence-sensitive work

    The build includes claims a real person may rely on.

  2. Direction

    Separate known from uncertain

    Ask for each claim, its source, source date, and what would disprove it. Prefer primary sources.

  3. Check

    Open the source yourself

    Confirm the source exists, says what the AI claims, and is current enough for the decision.

Copy, adapt, direct

List every checkable claim in this material. For each, give the best primary source, publication date, exact point it supports, and any uncertainty. If you cannot verify a claim, label it unverified instead of filling the gap.

Field panel 08

pattern

Pattern 06 - Bounded agent brief

Use when AI will take several steps toward a goal rather than answer once.

  1. Goal

    A finish condition

    Say what done means and what evidence the agent must return.

  2. Bounds

    Allowed, forbidden, approval

    Name tools it may use, actions it may not take, retry limits, and when a human must approve.

  3. Control

    Visible log and stop

    Require a step log, uncertainty flags, and a stop if inputs are missing or risk rises.

Copy, adapt, direct

Goal: [outcome]. Allowed tools: [list]. Never: [actions]. Ask approval before: [actions]. Stop when: [condition], after [limit] attempts, or when required data is missing. Return a step log, evidence, and unresolved risks.

Keep it in the field

Print it, mark it, return to the build.

The PDF carries the same source edition as this web guide and is designed with A4 and US Letter-safe print margins.

Download printable PDF