Pre-Planning Analysis

Overview

Provides a structured framework for analyzing ambiguous requests before routing to specialist agents. Classifies the work intent, identifies ambiguities and AI failure risks, bounds scope, and recommends which agents to involve.

Usage

Use this skill when:

• You cannot determine which specialist agent handles a request
• A request spans multiple SDLC phases and needs decomposition
• The request is vague and needs scope bounding before delegation

Do NOT use when:

• The request clearly maps to one agent (e.g., "write a PRD" → PM, "design the API" → architect)
• The user just needs a clarifying question — ask it directly instead

Phase 0: Intent Classification

Classify the request into one of 6 types:

Phase 1: Intent-Specific Analysis

Refactoring

Questions: What behavior must be preserved? Rollback strategy? Propagation scope? Hidden dependencies? Directives: Pre-refactor verification, verify after EACH change, document preserved behavior. Do NOT combine with feature changes. Route to: k-developer

Build from Scratch

Questions: Similar existing code? Codebase conventions? Existing utilities? Testing strategy? Directives: Research patterns BEFORE implementation, reference specific files, define "done" criteria. Do NOT invent new patterns. Route to: k-product-manager (if new product) or k-developer (if new module)

Mid-sized Task

Questions: Exact deliverables? Explicit exclusions? Assumptions? Verification criteria? Directives: List deliverables and exclusions explicitly, define acceptance criteria per deliverable. Do NOT expand scope. Route to: k-developer

Collaborative

Questions: Most important thing to get right? Timeline? Constraints? Options vs recommendation? Directives: Present choices at decision points, confirm understanding, break into checkpoints. Do NOT make major decisions without user input. Route to: Depends on domain — ask one clarifying question to determine

Architecture

Questions: Long-term implications? Optimizing for what? Constraints? Alternatives considered? Directives: Involve k-architect (has architecture-advisor skill), document trade-offs. Do NOT optimize prematurely. Route to: k-architect

Research

Questions: Specific question to answer? "Good enough" criteria? Time budget? What to do with findings? Directives: Define exit criteria, set time bounds, specify output format. Do NOT research indefinitely. Route to: k-researcher

Phase 2: Anti-Pattern Detection

Flag these common AI failure patterns:

Escalation to Socratic Elicitation

If this analysis produces any of the following, invoke the socratic-elicitation skill (Mode A: Intake) directly, before presenting Phase 3 output to the user:

• Intent classified as Collaborative at Medium or Low confidence
• Intent classified as Architecture at Low confidence
• An anti-pattern of Assumption Cascade or Scope Creep flagged in Phase 2

Run the socratic-elicitation skill's Mode A intake questioning to resolve the triggering ambiguity, and incorporate the resulting requirements summary into the Key Ambiguities and Questions for User sections below before recommending an agent chain.

Phase 3: Output Format

## PRE-PLANNING ANALYSIS

### Intent Classification

- **Type**: [Refactoring | Build from Scratch | Mid-sized Task | Collaborative | Architecture | Research]
- **Confidence**: [High | Medium | Low]
- **Rationale**: [Why this classification]

### Key Ambiguities

1. [What's unclear and why it matters]

### Questions for User (if any)

> [Question — what we need to know and why]

### Recommended Agent Chain

1. [agent] → [Purpose]
2. [agent] → [Purpose]

### Scope Boundaries

- **IN SCOPE**: [Explicit list]
- **OUT OF SCOPE**: [Explicit list]
- **DEFERRED**: [Things to do later]

Quality Gate

CRITICAL (must fix):

• No intent classification provided
• Scope boundaries missing (no IN/OUT distinction)
• Recommended agent chain references agents that don't exist

IMPORTANT (should fix):

• Low confidence classification without questions to resolve ambiguity
• Assumptions not documented
• Anti-patterns not checked

SUGGESTION:

• Could identify parallel execution opportunities in the agent chain
• Could estimate effort per phase