Idea Discovery Skill

Great products come from finding real problems before competitors do. This skill documents the three modes of proactive idea discovery and how they feed into the product lifecycle.

When to Activate

• You have user feedback (surveys, support tickets, reviews) to analyze
• You want to understand the competitive landscape in a space
• You need to generate ideas from scratch for a problem domain
• You're blocked on what to build next
• You want to validate that a problem is real before writing a PRD

---

The Discovery Pipeline

@startuml
!theme plain
skinparam backgroundColor #FAFAFA

rectangle "Discovery Sources" {
  rectangle "Inbound\n/analyze-feedback" as INB #E3F2FD
  rectangle "Outbound\n/discover" as OUT #E8F5E9
  rectangle "Creative\n/brainstorm" as CRE #FFF3E0
}

rectangle "Idea Capture\n/idea" as IDEA #F3E5F5

rectangle "Product Lifecycle" {
  rectangle "/evaluate" as EVAL
  rectangle "/explore" as EXP
  rectangle "/prd" as PRD
  rectangle "/overnight\nor /tdd" as IMPL
}

INB --> IDEA : "pain points\n+ idea seeds"
OUT --> IDEA : "market gaps\n+ idea seeds"
CRE --> IDEA : "idea seeds"
IDEA --> EVAL
EVAL --> EXP : "Go"
EVAL --> IDEA : "Modify"
EXP --> PRD
PRD --> IMPL
@enduml

---

Mode 1: Inbound — Analyze Feedback

Command: /analyze-feedback <file or paste>

Agent: feedback-analyst (model: opus)

When to use:

• You have real user feedback (support tickets, survey responses, NPS comments, App Store reviews, Intercom logs)
• You want to find patterns in what users complain about
• You want to extract the real job behind a feature request

What it does:

1. Semantic clustering — groups feedback by underlying pain, not surface keywords
2. JTBD extraction — turns complaints into "When [X], I want to [Y] so I can [Z]"
3. Pain intensity scoring 1-5
4. Opportunity Score = Frequency × Pain × (1 / Complexity)
5. Creates ranked pain points and idea seeds in docs/ideas/discovered/

Output: docs/insights/feedback-YYYY-MM-DD.md

Best for: Understanding what existing users actually suffer from. Ground truth over hypothesis.

---

Mode 2: Outbound — Competitive Research

Command: /discover <domain or topic>

Agent: competitive-analyst (model: opus)

When to use:

• Entering a new market or domain
• Checking if a problem space is already saturated
• Finding underserved segments competitors ignore
• Reading competitor strategic signals (job postings, changelogs, pricing changes)

What it does:

1. Researches 2-6 competitors systematically (homepage, pricing, changelog, G2/Reddit)
2. Builds a feature matrix (✓ full / ~ partial / ✗ missing)
3. Analyzes positioning: who they serve and what they emphasize
4. Identifies market gaps and underserved segments
5. Creates idea seeds from each gap

Output: docs/insights/discover-YYYY-MM-DD-<topic>.md

Best for: Finding what nobody does well. Gaps beat head-to-head competition.

---

Mode 3: Creative — Structured Ideation

Command: /brainstorm <problem space>

When to use:

• You have a domain but no data (early stage, pivoting, exploring)
• You want diverse ideas across multiple ideation angles
• You want to stress-test a problem space before committing to research

What it does — 4 frameworks in sequence:

Framework	Question	Output type
Jobs-to-be-Done	"What job is the user hiring a product to do?"	Functional need
How Might We	"HMW make X better?"	Solution direction
Analogy Thinking	"How does another industry solve this?"	Mechanism transfer
Constraint Reversal	"What if we removed [assumed constraint]?"	Breakthrough idea

Generates 15-25 raw ideas, filters to top 7 by novelty + JTBD fit + feasibility + market signal.

Output: 7 idea seeds in docs/ideas/discovered/

Best for: Early-stage exploration, creativity unblocking, pivots.

---

Choosing the Right Mode

Situation	Use
Have user feedback data	/analyze-feedback
Entering a competitive market	/discover
No data, need ideas fast	/brainstorm
Want all angles	Run all three, then /brainstorm with inputs from the others
Know the problem, need the solution	Skip to /idea → /evaluate

---

Idea Seeds

All three discovery modes produce idea seeds — small structured files saved to docs/ideas/discovered/:

# Idea: <name>

**Core insight:** <the non-obvious thing — one sentence>
**Job solved:** "When [X], I want to [Y] so I can [Z]"
**Different from existing:** <what no current solution does>
**Riskiest assumption:** <what must be true for this to work>
**Source:** <feedback analysis | competitive gap | brainstorm>
**Discovery date:** YYYY-MM-DD

These are inputs to /idea, which formalizes them and kicks off the lifecycle.

---

Full Discovery → Lifecycle Example

# Step 1: Discover ideas from multiple angles
/analyze-feedback docs/feedback/q4-survey.csv
/discover "developer error monitoring"
/brainstorm "help developers understand production errors faster"

# Step 2: Review idea seeds (in docs/ideas/discovered/)
# Pick the most compelling ones

# Step 3: Enter the product lifecycle
/idea smart-error-grouping
/evaluate smart-error-grouping
# → Go / No-Go / Modify

# Step 4: If Go
/explore smart-error-grouping
/prd smart-error-grouping

# Step 5: Implement
/overnight smart-error-grouping
# or: /tdd smart-error-grouping

---

Anti-Patterns

• Skipping discovery — building what seems obvious without checking user data or the competitive landscape
• Analyzing too long — discovery is an input, not the product. Cap at 2-3 sessions per cycle
• Over-indexing on one mode — feedback tells you about existing users; competitive tells you about the market; brainstorm tells you what's possible. Use all three
• Treating idea seeds as commitments — they're hypotheses. Most won't survive /evaluate
• Copying competitors — competitive research surfaces gaps, not features to clone