Strategy

Structured thinking for brainstorming, prompt engineering, and decision making.

Workflow

Step 1: Intake & Framing

Extract the user's core need and classify it into one of three tracks:

Track	Trigger	Output
Decision	"Should I...", "Which option...", "Help me choose..."	Scored recommendation with rationale and runner-up tradeoff
Ideation	"Brainstorm...", "Generate ideas for...", "What are ways to..."	Prioritized list of ideas with convergent selection
Clarity	"How do I think about...", "Improve this prompt", "Reframe..."	Restructured framing, 7-dimension prompt, or first-principles breakdown

If the user's need spans multiple tracks, run them sequentially: Clarity → Ideation → Decision.

Step 2: Select Framework

For Decision track: Choose ICE if speed matters, Impact/Effort matrix if resources are constrained, First Principles if the problem feels stuck or assumed, Pre-Mortem if the decision is high-stakes and irreversible.

For Ideation track: Start with brainwriting (quantity). If ideas stall, inject SCAMPER or reverse brainstorming. Never open with free association for groups of 3+ — dominant voices take over.

For Clarity track: Run the 7-dimension prompt diagnosis. Identify which dimensions are weak (0-3 on a 1-10 scale). Rewrite the prompt addressing the weakest dimensions first.

Step 3: Execute Framework

Follow the framework protocol from its dedicated section below. Timebox strictly:

• Decision: 20 min (5 min framing, 10 min scoring, 5 min decision)
• Ideation: 60 min (divergent 30 min, cluster 10 min, convergent 20 min)
• Clarity: 10 min (diagnose 3 min, rewrite 5 min, changelog 2 min)

Step 4: Document & Deliver

Every strategy output must include:

• The chosen framework and why it was selected
• The raw output (scores, ideas, rewritten prompt)
• One recommended action with owner + next step + deadline
• One explicitly rejected alternative and the reason

Step 5: Verify Against Production Checklist

Run the Production Checklist before delivering. If any item is unchecked, return to the relevant step. Never deliver a strategy artifact that skips convergent selection or lacks a specified output format.

Brainstorming

1. Process

1. Divergent (generate, no judgment) — 15-30 min
2. Clustering (organize, group, eliminate) — 10-15 min
3. Convergent (select, vote, prioritize) — 15-20 min

Never mix divergent and convergent — judgment kills ideas before they form.

2. Divergent Techniques

Technique	How	Best for
Free association	Write everything, no filter	Warm-up, quantity
Brainwriting 6-3-5	6 people write 3 ideas in 5 min, pass	Avoiding dominant voices
Reverse brainstorming	How to make it worse? Then reverse.	Breaking assumptions
Random word	Random noun → force connections	Lateral thinking
SCAMPER	Substitute, Combine, Adapt, Modify, Put to use, Eliminate, Reverse	Systematic exploration
HMW	Reframe as "How Might We" questions	Design thinking
Worst idea	Generate terrible ideas, then invert	Reducing fear

3. Convergent Techniques

Technique	How
Dot voting	3-5 votes per person
Impact/effort matrix	X: impact, Y: effort
ICE scoring	Impact, Confidence, Ease (1-10, average)
NUF test	New, Useful, Feasible (1-10 each)

4. Session Template

Phase	Duration	Activity
Problem framing	10 min	Define "How might we..." question
Warm-up	5 min	Low-stakes exercise
Divergent round 1	15 min	Brainwriting. Target: 30+ ideas
Cluster	10 min	Affinity mapping
Divergent round 2	10 min	SCAMPER on strongest clusters
Convergent	15 min	Impact/effort matrix. Top 3-5
Action planning	10 min	Owner, next step, deadline

5. Session Ground Rules

• Defer judgment. Go for quantity (set targets: "50 ideas in 20 min")
• One conversation at a time. Build on "Yes, and..."
• Encourage wild ideas. Round-robin or brainwriting.
• Hard stop at 90 min.

Decision Frameworks

ICE Scoring

Score = Impact (1-10) × Confidence (1-10) × Ease (1-10)
Higher = prioritize first

Impact/Effort Matrix

           High Impact          Low Impact
Low Effort  → Do First         → Quick Wins
High Effort → Strategic        → Don't Do

First Principles Thinking

1. Identify the current belief/assumption
2. Break it down into fundamental truths
3. Rebuild from those truths

Current: "Our testing takes too long to run"
Deconstruction: What IS testing? Verifying code behavior.
Fundamental: We need confidence code works. Speed matters but accuracy matters more.
Rebuild: What's the fastest way to get accuracy confidence? Change test scope to integration-focused.

Pre-Mortem Analysis

Before starting a project, imagine it failed 6 months from now:

1. What went wrong? (list 5-10 failure modes)
2. For each, what's the probability? (low/medium/high)
3. For each, what's the impact? (minor/major/critical)
4. Mitigation: What can we do NOW to prevent it?

Opportunity Cost

Choosing Option A means NOT choosing Options B-Z.
Before committing, list: What else could these resources do?

Prompt Engineering

The 7 Dimensions

1. Clarity — Remove ambiguity. Every noun and verb should have one interpretation.
2. Specificity — Replace vague language with concrete details. Use names, paths, numbers.
3. Constraints — Technology, resource, time, compliance hard boundaries.
4. Format — Bullet list, table, code block, JSON, Mermaid. Tell the model how to answer.
5. Examples — One example is worth 100 words of description.
6. Tone — Internal message (direct), client email (professional-cordial), docs (neutral-precise), bug report (factual-minimal).
7. Iteration Signal — "Start here, refine later" vs "Production-ready" vs "Give me options" vs "Just the code"

Enhancement Workflow

1. Diagnose: Run through 7 dimensions — identify which are weak or missing
2. Rewrite: Complete rewrite, don't annotate original
3. Changelog: 1-2 bullet points per dimension improved

Specificity Table

Vague	Specific
"help with this code"	"refactor auth middleware in src/middleware/auth.ts"
"some options"	"exactly 3 options ranked by cost"
"make it better"	"improve load time from 3s to <1s on mobile Chrome"
"explain"	"explain in 3 bullet points to a junior dev"

Error Handling

Error	Cause	Resolution
Session runs over 90 min	Poor timeboxing	Assign a timer. Hard stop at 90 min regardless of progress
Dominant voice dominates	No facilitation structure	Switch to brainwriting or round-robin. No open floor until last 10 min
Ideas too similar / not novel	Groupthink	Insert reverse brainstorming or random-word technique mid-session
No clear decision after convergent phase	Tie votes	Use ICE scoring as tiebreaker. Default to impact if still tied
Prompt still produces bad output after rewrite	Diagnosed wrong dimension	Re-run 7-dimension check. Most often: missing constraint or missing example
"I don't have ideas" block	Blank-page problem	Use worst-idea technique or reverse brainstorming to unstick

Production Checklist

Before delivering a strategy artifact (session output, prompt, decision document):

• Problem statement defined in "How might we..." format
• Divergent and convergent phases kept separate with a clear transition
• Convergent selection uses at least one scoring method (ICE, impact/effort, dot voting)
• Prompt covers all 7 dimensions — no dimension is "N/A" without reason
• Prompt includes at least one example (dimension 5)
• Output format specified in prompt (dimension 4)
• Iteration signal included (dimension 7)
• Decision documented with rationale and runner-up tradeoff
• Action items: owner + next step + deadline

Anti-Patterns

Anti-Pattern	Why It Fails	Correction
Mixed divergent and convergent	Judgment kills idea generation before it starts	Separate phases with a break. No "yes but" during divergent
No warm-up	Cold start produces safe, boring ideas	Always do 5-min warm-up (e.g. "ideas for improving a paperclip")
Dominant voices take over	Junior or quiet team members disengage	Use brainwriting, round-robin. No open discussion until convergent
Too many people (8+)	Diffusion of responsibility, not enough airtime	4-8 max per session. Split into parallel groups if larger team
No follow-through	Great ideas die after the session	Assign ownership, next step, and deadline before closing
Prompt too short (<10 words)	Model fills gaps with wrong assumptions	Expand to cover context, constraint, format, and example
Compound request	Model optimizes for first instruction only	Split into separate prompts or use numbered list with weighting
No output format	Model chooses — rarely the most useful one	Always specify format (JSON, table, bullet list, code block)
Using ICE without confidence check	False precision — confidence is often guessed	Calibrate confidence: "how sure are you on a 1-10?" and average across 3 people
Skipping opportunity cost	Overcommit to first good option	Before finalizing, ask: "What else could these resources do?"

Advanced Decision Frameworks

RICE Scoring

Score features: Score = (Reach * Impact * Confidence) / Effort

• Reach: how many users affected (1-1000+)
• Impact: 0.25 (minimal), 0.5 (low), 1 (medium), 2 (high), 3 (massive)
• Confidence: 20% (gut), 50% (some data), 80% (quantified), 100% (A/B tested)
• Effort: person-months

Kano Model

Classify features: Basic (expected, absence = dissatisfaction), Performance (more = better), Delighter (unexpected, absence = neutral). Prioritize: Basic → Performance → Delighter.

OKR-to-Task Decomposition

• Objective: qualitative, inspiring, time-bound (quarter)
• Key Results: 3-5 quantitative outcomes
• Tasks: decompose each KR into 2-3 actionable tasks per week

Sources

• Alex Osborn "Applied Imagination"
• IDEO design thinking
• d.school Stanford facilitation guides
• Jake Knapp "Sprint"
• Annie Duke "Thinking in Bets"
• First Principles (Elon Musk / Richard Feynman approach)
• Fermi Estimation techniques

Checklist

• Skill loads without errors in the AI agent
• YAML frontmatter is valid (description, compatibility, audience)
• Workflow section provides clear step-by-step instructions
• Error handling section covers common failure modes
• All referenced files (references/, scripts/, assets/) exist
• Skill triggers correctly for intended use cases
• No broken links or missing resources