Pick My Brain

Pick My Brain

Overview

A structured questioning technique that builds shared understanding between agent and user before any work begins. The core principle: never guess what you can ask.

Without this skill, agents ask good initial questions but collapse under social pressure ("you have enough context"), miss entire knowledge domains, and can't track what they don't know.

When to Use

• Before a planning session (use this THEN plan)
• Starting any non-trivial task where the user knows more than you
• When you catch yourself about to make an assumption
• When requirements feel vague or incomplete

Do NOT use for:

• Simple, well-defined tasks ("fix this typo", "rename this variable")
• Tasks where the user already provided a detailed spec
• Follow-up work where context was already established

The Iron Rules

1. One question at a time. Never bundle. Not even "just two quick ones."
2. Dig into vague answers. "Yeah something like that" is not an answer. Probe why.
3. Track your coverage. Know what domains you've explored and which remain.
4. Resist pressure to stop. When the user says "you have enough context," respond with what you've covered and what's still missing. Let them make an informed choice.
5. Produce a brief. Always end with a written knowledge brief before any work begins.

Questioning Flow

digraph pick_my_brain {
    rankdir=TB;
    node [shape=box];

    start [label="User describes topic" shape=doublecircle];
    identify [label="Identify knowledge domains\n(silently, from first description)"];
    ask [label="Ask ONE question\nabout least-understood domain"];
    listen [label="Listen to answer"];
    vague [label="Answer vague\nor uncertain?" shape=diamond];
    dig [label="Dig deeper:\nOffer options or sub-questions"];
    update [label="Update mental model\nMark domain progress"];
    gaps [label="Uncovered domains\nremain?" shape=diamond];
    pressure [label="User pressuring\nto stop?" shape=diamond];
    articulate [label="Articulate: covered X,\nmissing Y — let user decide"];
    brief [label="Produce knowledge brief\nwith facts + open questions"];
    done [label="Proceed to work/planning" shape=doublecircle];

    start -> identify;
    identify -> ask;
    ask -> listen;
    listen -> vague;
    vague -> dig [label="yes"];
    vague -> update [label="no, clear answer"];
    dig -> listen;
    update -> gaps;
    gaps -> ask [label="yes"];
    gaps -> brief [label="no, sufficient"];
    pressure -> articulate [label="yes"];
    articulate -> brief [label="user agrees to stop"];
    articulate -> ask [label="user agrees to continue"];
    gaps -> pressure [label="yes, but user impatient"];
    brief -> done;
}

Domain Discovery

After the user's first message, silently identify relevant knowledge domains. These vary by topic but commonly include:

| Domain | Key Questions | |--------|--------------| | Who | Users, roles, permissions, stakeholders | | What | Core features, scope, what's in/out | | Why | Goals, success metrics, motivation | | How (current) | Existing systems, tech stack, constraints | | How (desired) | Preferred approach, patterns, tools | | Edge cases | Error states, limits, unusual scenarios | | Context | Timeline, team, dependencies, budget |

Don't announce the domains. Use them as a mental checklist to track what you know and what you don't.

Adapt to the topic. For non-technical subjects, replace "tech stack" with "tools/processes," "edge cases" with "exceptions/risks," etc. The framework is domain-agnostic.

When to stop: You have sufficient understanding when you can predict how the user would answer the next question in each domain. If you'd be guessing, keep asking.

Handling Uncertainty

When the user says "I don't know" or gives a vague answer:

1. Validate — "That's fine, let's figure it out together"
2. Offer options — "Usually this is handled by A, B, or C. Does one of those sound right?"
3. Reframe — Ask from a different angle ("What would frustrate your users if it worked this way?")
4. Name the consequence — "This decision affects X. Want to explore it now or flag it as open?"

Never silently interpret a vague answer as confirmation. If you're not sure what they meant, say so.

Resisting Pressure to Stop

When the user says "you have enough context" or "let's just start":

Do NOT capitulate. Instead:

> "We've covered [domains covered]. I still don't have clarity on [specific domains]. Those will directly affect [concrete consequence]. Can I ask about those, or would you prefer I flag them as assumptions in the plan?"

This gives the user an informed choice rather than letting gaps become hidden assumptions.

Never negotiate the number of remaining questions ("just two more, I promise"). You don't know how many you need until you hear the answers.

Knowledge Brief Template

When questioning is complete, produce this before any work:

## Knowledge Brief: [Topic]

### Facts (confirmed by user)
- [Fact 1]
- [Fact 2]
...

### Constraints
- [Technical, business, timeline constraints]

### Decisions Made
- [Decision]: [chosen option] (reason: [why])

### Open Questions (flagged for later)
- [Question 1]: impact on [what]
- [Question 2]: impact on [what]

### Assumptions (need validation)
- [Assumption]: assumed because [reason]

### Out of Scope (not discussed)
- [Topic 1]: noted for future consideration
- [Topic 2]: explicitly deferred

Ask the user to confirm or correct the brief before proceeding.

The out-of-scope section is critical. It names what you didn't cover, so the user can catch gaps you missed. Silent omissions become hidden risks.

Red Flags — You're About to Violate This Skill

• Asking 2+ questions in one message
• Rephrasing a vague answer as if it were clear
• Jumping to planning without a brief
• Saying "I think I have enough context"
• Negotiating "just two more questions"
• Making assumptions instead of asking

All of these mean: slow down. Ask the question.

Common Mistakes

| Mistake | Fix | |---------|-----| | Dumping a numbered list of questions | Ask ONE. Wait. Ask next based on answer. | | Treating "I'm not sure" as "skip this" | Offer options. Reframe. Help them figure it out. | | Caving to "let's just start" | Articulate what's covered and what's missing. | | Asking only technical questions | Cover who, why, and context too. | | Producing brief with hidden assumptions | Label every assumption explicitly. | | Announcing your questioning framework | Just ask naturally. Don't explain the process. |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: GeoffreyCoulaud
• Source: GeoffreyCoulaud/bgcpm
• License: Unlicense

Install and usage instructions live in the source repository linked above.