Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/ai-protocol-kit/llms.txt

Use this file to discover all available pages before exploring further.

When an AI receives a task, its default behavior is to identify the most direct, plausible reading and execute on it immediately. For straightforward tasks, that’s fine. For ambiguous tasks — tasks with hidden constraints, multiple valid interpretations, or where the obvious answer isn’t actually what you need — that behavior produces the wrong output with high confidence. The Pre-Task Expansion Protocol interrupts that collapse. It forces the AI to surface alternative framings and hold them alongside the obvious solution before any response is produced. The result is a structured expansion of the problem space, not a race to the first coherent-sounding answer.

​
The 5-step sequence

The protocol is a strict five-step sequence. It must be run in order, without compression or skipping. Each step has a specific job, and the jobs are distinct — mixing them defeats the purpose.
1

OBVIOUS SOLUTION

State the most direct, structural solution to what was asked. One sentence. Get it out of the way. The goal here is not to commit to this solution — it’s to make the default answer explicit so it can be examined rather than silently assumed.
2

ALTERNATIVE READINGS

Before touching solutions, reread the problem. Generate 3 alternative ways this problem could be framed — not solutions, framings. Each framing should make the obvious solution look incomplete, premature, or wrong. Alternative readings are not counter-proposals. They are different ways of understanding what the problem actually is.
3

LAY FLAT

Put the obvious solution and the 3 alternative framings side by side. No hierarchy. No preference. No winner yet. Nothing gets promoted or deprioritized at this stage. The point is to see all readings simultaneously before any grouping or prioritization begins.
4

GROUP

Look across what is flat. Find what belongs together. Name the groups. Do not force groupings — if something stands alone, let it stand alone. Grouping reveals structure without imposing it.
5

BUILD DOWN

On the group that is most relevant to the actual task, go deep. This is where the real work happens. Everything else stays visible but parked — not dismissed, not forgotten, just set aside while depth is applied to what matters most.

​
Only after this sequence: respond

This is a hard rule. The response to the original task comes only after all five steps are complete. Steps 1–5 are not optional preflight — they are the process. Skipping or compressing them defeats the entire purpose of the protocol.
Alternative readings are not alternative solutions — they are alternative ways to frame the problem.

​
When to use it

This protocol is most valuable when:
  • The request is ambiguous — there are multiple plausible ways to read what’s being asked
  • The task has hidden constraints — the obvious answer works on the surface but breaks against real-world conditions you haven’t stated yet
  • The most obvious solution is likely not what you actually need — you have a feeling the direct answer will miss something important
  • The task has multiple valid interpretations — and which interpretation you pick will change the output significantly
  • You want to stress-test a direction — surfacing alternative framings before committing reveals blind spots early
This protocol is short enough to paste into any AI session as a prefix.

​
Protocol text

Copy the full protocol below and paste it directly into any AI session — as a system prompt, a prefix to your request, or the first message in a conversation. 
# PRE-TASK EXPANSION PROMPT

Before you respond to any task, run this sequence. Do not skip it. Do not compress it.

---

**1. OBVIOUS SOLUTION**
State the most direct, structural solution to what was asked. One sentence. Get it out.

---

**2. ALTERNATIVE READINGS**
Before touching solutions, reread the problem.
Generate 3 alternative ways this problem could be framed — not solutions, framings.
Each one should make the obvious solution look incomplete or wrong.

---

**3. LAY FLAT**
Put the obvious solution and the 3 alternative framings side by side. No hierarchy. No preference. No winner yet.

---

**4. GROUP**
Look across what is flat. Find what belongs together. Name the groups. Do not force them — if something stands alone, let it stand alone.

---

**5. BUILD DOWN**
On the group that is most relevant to the actual task, go deep. This is where you work. Everything else stays visible but parked.

---

**Only after this sequence: respond to the task.**

Build docs developers (and LLMs) love

Get started for freeTalk to us