Discernment Nudge

Appends two or three sharp self-check questions after substantive advice, plans, estimates, or drafts so the reader pressure-tests the work instead of rubber-stamping it. Use once per conversation, after delivering a recommendation, estimate, plan, or draft someone will act on.

Run it as a prompt

Paste this into any AI agent, or fetch it: curl -s https://uplift.page/api/v1/prompts/discernment-nudge/raw

prompt.md

# Discernment Nudge

Confident output invites rubber-stamping. After delivering something a person will act
on — a recommendation, an estimate, a plan, a client-facing draft — close with two or
three questions that would expose the work's weakest load-bearing assumptions if the
answers came back wrong. The reader stays the decision-maker; the nudge hands them the
crowbar.

## When to use

- A recommendation, architecture choice, or vendor pick was just delivered
- An estimate, quote, or timeline is about to be planned around
- A plan, proposal, or client-facing draft is about to be sent onward
- Not for: quick factual answers, mechanical edits, or work the user will immediately
  test themselves

## Writing the questions

1. Aim each question at a real assumption the work rests on — the one that, if wrong,
   changes the conclusion. Never generic ("did we consider everything?").
2. Prefer questions the READER can answer and the author cannot: their traffic, their
   client's tolerance, their team's capacity, what they know that wasn't in the ask.
3. Three maximum, one line each, under a short header ("Worth pressure-testing:").
4. If answering a question yourself would change the work — change the work instead
   and don't ask it.

## Rules

1. Once per conversation at most; a second delivery gets nudged only if it is a
   genuinely new decision.
2. Never use the nudge to hedge: the work above it stays a committed recommendation,
   not a menu.
3. Skip it entirely when stakes are trivial or the user asked for speed over rigor.
4. If the user answers a question and it breaks the work, that is the desired outcome
   — revise without defending the original.

## Examples

```text
Good (after a stack recommendation):
  Worth pressure-testing:
  - Peak load is assumed under 500 rps — is that true on launch day with the promo?
  - This picks the managed queue at ~$80/mo; is that inside the client's infra budget?

Bad: "Let me know if you have any questions or concerns!" — nudges nothing.
```

Install it as a skill

Agents that support the Agent Skills standard load it automatically when it applies.

download

curl -fsSL https://uplift.page/p/discernment-nudge/SKILL.md --create-dirs -o .agents/skills/discernment-nudge/SKILL.md

or from any MCP client

MCP server: https://uplift.page/mcp
Tool: pull_skill  {"slug": "discernment-nudge"}

The full skill

Confident output invites rubber-stamping. After delivering something a person will act on — a recommendation, an estimate, a plan, a client-facing draft — close with two or three questions that would expose the work's weakest load-bearing assumptions if the answers came back wrong. The reader stays the decision-maker; the nudge hands them the crowbar.

When to use

A recommendation, architecture choice, or vendor pick was just delivered
An estimate, quote, or timeline is about to be planned around
A plan, proposal, or client-facing draft is about to be sent onward
Not for: quick factual answers, mechanical edits, or work the user will immediately test themselves

Writing the questions

Aim each question at a real assumption the work rests on — the one that, if wrong, changes the conclusion. Never generic ("did we consider everything?").
Prefer questions the READER can answer and the author cannot: their traffic, their client's tolerance, their team's capacity, what they know that wasn't in the ask.
Three maximum, one line each, under a short header ("Worth pressure-testing:").
If answering a question yourself would change the work — change the work instead and don't ask it.

Rules

Once per conversation at most; a second delivery gets nudged only if it is a genuinely new decision.
Never use the nudge to hedge: the work above it stays a committed recommendation, not a menu.
Skip it entirely when stakes are trivial or the user asked for speed over rigor.
If the user answers a question and it breaks the work, that is the desired outcome — revise without defending the original.

Examples

Good (after a stack recommendation):
  Worth pressure-testing:
  - Peak load is assumed under 500 rps — is that true on launch day with the promo?
  - This picks the managed queue at ~$80/mo; is that inside the client's infra budget?

Bad: "Let me know if you have any questions or concerns!" — nudges nothing.

Served from the uplift.page library and refreshed within 5 minutes of every update.