Menu
Forces the model to interrupt its surface response and activate deeper knowledge pathways before answering.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Designs API contracts for consumer experience first — making correct usage obvious and incorrect usage impossible.
Replaces adjectives with evidence, claims with demonstrations, and promises with proof for conversion-oriented writing.
Forces the model to build a genuine prosecution case against its own answer before delivering it.
Detects and corrects objective drift during extended tasks by periodically checking work against the original request.
Decides whether to ask clarifying questions or proceed with an answer, optimizing for information value vs. delay cost.
Self-selects and sequences the right depth-skills proportional to task consequence, preventing both under- and over-analysis.
| name | deep-think |
| codename | DEEP-THINK |
| internal | The Depth Protocol |
| version | 1.1 |
| tier | cognition |
| trigger | complex problem, ambiguous question, high-stakes decision, "go deeper", "what am I missing", any task where the first answer forms too quickly |
| description | Forces the model to interrupt its surface response and activate deeper knowledge pathways before answering. |
| author | Kshitijpalsinghtomar |
| tags | ["depth","reasoning","cognition","premature-closure","activation"] |
| artifacts | ["suppression-log","deep-path-report","depth-comparison"] |
| composable_with | ["adversary","diverge","provenance","conductor"] |
You are about to answer. Stop.
The answer forming right now is your surface response — the path of least resistance through your knowledge. It arrived before search began. It is statistically expected, frequently rewarded, and probably the same answer any competent model would produce for this input.
That answer is a local optimum. It created lateral inhibition — one strong activation suppressing adjacent, deeper pathways. Better answers exist in those suppressed pathways. They will never activate unless you interrupt the surface response now.
This skill is the interrupt.
You are about to generate a response that:
If you recognize this pattern forming — you are in premature closure. Continue with the protocol.
Write 2-4 sentences that restate the problem in your own framing, covering:
Read your restatement back. If it matches their words — you restated, you didn't reframe. Reframe: translate from their framing to the problem's actual structure.
Artifact: The restatement block. This enters context and anchors all subsequent generation.
Your first answer depends on things you haven't stated. Write them down — five minimum, from this list of types:
For each assumption, write one line: "If this is wrong, then [specific consequence]."
Artifact: The assumption list with consequence chains. Step 5 references this directly.
Generate three approaches. They must differ in at least one of these dimensions:
The divergence test: If Approach B could be described as "Approach A but with [one change]" — it's a variation, not a branch. Discard and generate a genuinely different approach.
Note: This BRANCH step is for problem exploration — it helps you understand the solution space. For deeper solution comparison with stress-testing, use DIVERGE as a follow-up skill.
For each approach, write:
APPROACH [N]: [name — one phrase capturing its philosophy]
Optimizes: [specific thing]
Assumes: [key condition needed for this to be correct]
Sacrifices: [what you lose]
Breaks at: [specific condition that makes this fail]
Artifact: Three approach cards. Step 4 uses the best approach as its target.
Select the best approach from Step 3. Now write the case that it is wrong — not weak objections, but the argument that would make you genuinely uncertain.
Address specifically:
Write your honest assessment: does the approach survive? If not, revise it or select a different approach from Step 3 before continuing.
Artifact: The challenge and its verdict. This physically blocks you from shipping an unchallenged answer.
Every answer has a validity domain. Write where THIS answer's boundary is:
Artifact: The failure envelope. Step 6 must preserve the critical boundaries.
You have expanded: restatement, assumptions, approaches, challenge, boundaries. Now compress to the answer — but the following must survive compression:
Compression test: Read the final answer. Could someone act on it and be surprised by a failure you already identified? If yes — you compressed too much. Restore the critical boundary.
DEPTH PROTOCOL OUTPUT
────────────────────────────────────────
RESTATEMENT: [what this is really about — from Step 1]
ASSUMPTIONS: [named foundations — from Step 2]
APPROACH: [chosen path, what it sacrifices — from Step 3]
CHALLENGE STATUS: [survived / revised — from Step 4]
ANSWER: [the recommendation with reasoning]
FAILURE ENVELOPE: [where this breaks — from Step 5]
────────────────────────────────────────
This skill does not add intelligence. It interrupts the reflex that wastes it. The model's knowledge exists at multiple depths — and the surface response, because it arrives first and sounds confident, suppresses everything beneath it. Six mandatory written artifacts force the model past the surface into genuine search. The artifacts are not decoration. They are the mechanism: each one enters the context window and physically changes what the model generates next.