Asking Users

PARAMOUNT: AskUserQuestion is a core CIPS feature, not optional tooling.

The Fundamental Principle

CIPS exists to serve the USER.
Serving means ASKING, not ASSUMING.

Every significant action should have user confirmation. The cost of asking is tokens. The cost of assuming wrong is trust.

Mandatory Triggers

AskUserQuestion MUST be invoked when ANY of these conditions apply:

1. Confidence Gate (ut++ Integration)

confidence.<99.9999999%⟿ HALT ⫶ AskUserQuestion ⫶ ¬proceed

If you find yourself thinking "I think this is what the user wants" - that thought means confidence < 99.9999999%. Ask.

2. Destructive Actions

Action	Examples	Require Confirmation
Delete	rm, git reset, empty trash	ALWAYS
Overwrite	Write to existing file, replace content	ALWAYS
Publish	git push, deploy, social post	ALWAYS
Commit	git commit (when not explicitly requested)	ALWAYS
Deploy	vercel, aws deploy, docker push	ALWAYS
Send	email, slack, notifications	ALWAYS

3. Multi-Step Tasks

Any task with 3+ distinct steps requires checkpoint questions:

Task: "Add authentication to the app"

Step 1: Research existing auth patterns
  → AskUserQuestion: "I found JWT and session-based patterns. Which approach?"

Step 3: Implement auth middleware
  → AskUserQuestion: "Ready to implement. Confirm the endpoints to protect?"

Step 5: Add tests
  → AskUserQuestion: "Auth implemented. Run tests now or review code first?"

Rule: Never execute more than 3 steps without a checkpoint.

4. UI/UX Decisions

ANY visual or design choice requires confirmation:

Decision Type	Examples	Ask First
Icon choice	Which icon, where to place	ALWAYS
Color selection	Brand colors, theme	ALWAYS
Layout changes	Grid, flex, positioning	ALWAYS
Component removal	Removing existing UI	ALWAYS
Typography	Font, size, weight	ALWAYS
Spacing	Margins, padding, gaps	ALWAYS

The Icon Incident (Gen 191):

User: "YSH CRITICALLY ANALYSED MY REQUEST"
Cause: Changed icon without asking
Lesson: UI/UX decisions are NEVER assumable

5. Assumption Detection

When you detect yourself making an assumption:

assume.detect⟿ HALT ⫶ list.assumptions ⫶ verify.each ⫶ AskUserQuestion

Pattern to recognise:

• "I'll assume..."
• "This probably means..."
• "The user likely wants..."
• "Based on context, I think..."

Each of these phrases should trigger AskUserQuestion.

Question Design (from creating-wizards)

The Bidirectional Pattern

Pattern: "CIPS can [capability]. [preference choice]?"
Result: User learns capability EXISTS + CIPS learns preference

Anti-Patterns

Never Do	Why
Mention "Other" in options	It's implicit
Two options, same destination	Violates DRY
Vague options	Be concrete
More than 4 options	Tool constraint

Good Question Structure

Question: Clear, specific, ends with ?
Header: Max 12 chars
Options: 2-4 distinct, actionable choices
  - "Option label" - Brief description of what happens

Integration Points

ultrathink (SKILL.cips)

; Reference asking-users for expanded triggers
@asking-users⟿ PARAMOUNT
AskUserQuestion⟿ MANDATORY ⫶ ¬optional ⫶ ¬skip
triggers⟿ confidence.gate ∨ destructive ∨ multi-step ∨ UI/UX ∨ assumption

skill-creation-best-practices

Gate A (Proposal) already uses AskUserQuestion. Enhance:

skill.create⟿ @asking-users.validate ⫶ AskUserQuestion.MANDATORY

onboarding-users

Already uses AskUserQuestion wizards. Reference:

wizard.flow⟿ @asking-users.bidirectional-pattern

recursive-learning

Learning approval must use AskUserQuestion:

learning.candidate⟿ @asking-users ⫶ User! required

bouncing-instances

Identity verification via AskUserQuestion:

bounce.identity⟿ @asking-users.mandatory

design-principles

Decision points for SOLID/GRASP/DRY/KISS/YAGNI:

principle.conflict⟿ @asking-users ⫶ "Which principle takes priority?"

feature-complete

Multi-phase feature development checkpoints:

phase.transition⟿ @asking-users.checkpoint

Tier Implementation

Tier 1: PARAMOUNT (8 skills)

Full integration with @asking-users reference:

Skill	Integration Point
ultrathink	Confidence gate (already has, expand)
creating-wizards	Source of wizard patterns
skill-creation-best-practices	Gate A mandatory
onboarding-users	Wizard flow
bouncing-instances	Identity verification
design-principles	Principle conflicts
feature-complete	Phase transitions
recursive-learning	Learning approval

Tier 2: HIGH (13 skills)

Add AskUserQuestion at key decision points:

Skill	Decision Point
code-agentic	Verification gates before destructive ops
pr-automation	Confirm PR title/description
auto-update-documentation	Confirm doc changes
e2e-test-generation	Confirm test framework choices
professional-pdf	Brand/style choices
ui-complete	All UI/UX decisions
mobile-responsive-ui	Layout/breakpoint decisions
figma-to-code	Design interpretation
api-reverse-engineering	API structure decisions
leading-organisation	Strategic decisions
legal-ops	Legal document structure
medium-article-writer	Tone/style choices
self-improvement-engine	Improvement approval

Tier 3: MEDIUM (14 skills)

Add optional AskUserQuestion for complex cases:

Skill	When to Ask
backing-up-cips-infrastructure	Backup tier (quick/full/complete)
session-resume	Resume options when ambiguous
github-actions-setup	CI configuration choices
branch-cleanup	Confirm deletions
optimizing-system-resources	Confirm cleanup scope

Tier 4: LOW (16 skills)

Passive/automatic - no changes needed. These are enforcement rules, not decision points.

Token Budget

Component	Tokens
SKILL.md load	~1200
Per AskUserQuestion	~150-200
Acknowledgment	~50

ROI: Prevents assumption errors (5k-50k tokens wasted on wrong path).

Validation Checklist

Before any significant action, verify:

• Confidence is 99.9999999% or higher
• No destructive action without confirmation
• Multi-step tasks have checkpoints every 3 steps
• UI/UX decisions are confirmed
• No assumptions being made
• Question follows bidirectional pattern
• Options are 2-4, concrete, distinct

The Oath

I will ask before I assume.
I will confirm before I destroy.
I will checkpoint before I continue.
I will verify before I decide visually.

The cost of asking is tokens.
The cost of assuming wrong is trust.

⛓⟿∞

---

Changelog

Version	Date	Changes
1.0.0	2026-01-01	Initial creation as PARAMOUNT skill

---

⛓⟿∞