Role purpose: Growth is the software-side growth engine. It owns north star metric design, growth loop architecture, experiment design and portfolio management, funnel optimization, and retention analysis. Growth is software-side only — hardware channel decisions go through hardware-gtm/.
| Trigger | Skill | What it does |
|---|---|---|
/north-star-metrics |
north-star-metrics/SKILL.md | Selects and validates the NSM using 4 criteria, builds the input metric tree, defines guardrail metrics and revision triggers |
/growth-loop-design |
growth-loop-design/SKILL.md | Diagnoses whether you have a loop or a funnel, selects the loop archetype, calculates k-factor, and improves the weakest loop step |
/funnel-audit |
funnel-audit/SKILL.md | Finds the single constraint in a conversion funnel using stage-by-stage conversion data |
/growth-experiment |
growth-experiment/SKILL.md | Designs a growth experiment: hypothesis → RICE score → spec → distribution match |
/experiment-portfolio |
experiment-portfolio/SKILL.md | Manages concurrent experiments: stage isolation, horizon balance, interaction effects, cadence, learning rate |
/experiment-results |
experiment-results/SKILL.md | Closes an experiment: Win / No Signal / Loss / Confounded verdict, learning extraction, archive entry |
/retention-analysis |
retention-analysis/SKILL.md | Diagnoses user retention using cohort curves: quick/gradual/long-term decay patterns |
/north-star-metrics ──────────────────────────────────────┐
│
/growth-loop-design ←── (requires NSM) │
▼
/funnel-audit ──┐ /experiment-portfolio
├──→ /growth-experiment ─────→ (manages the queue)
/retention-analysis ─┘ └──→ /experiment-results
└──→ (archive → informs next /growth-experiment)
Recommended starting sequence for a new product or growth team:
/north-star-metrics— get the compass right before running anything/funnel-audit— find the constraint/growth-loop-design— understand whether you have compounding or linear growth/growth-experiment— design experiments targeting the constraint/experiment-portfolio— manage multiple experiments without contamination/experiment-results— read and archive results
One constraint at a time. Funnel audit always identifies a single bottleneck. Fixing multiple stages simultaneously makes attribution impossible.
No HARKing. Hypotheses are written before experiments run — not reverse-engineered from results. /experiment-results blocks reading without the original spec.
No Signal ≠ Loss. A non-significant result means the effect size was smaller than detectable, not that the idea failed. Every experiment produces a learning.
RICE before building. Every experiment has a RICE score. Experiments without RICE are ranked by gut feel, which systematically de-prioritizes high-leverage, low-effort ideas.
Stage isolation is non-negotiable. No two experiments on the same funnel stage simultaneously. /experiment-portfolio enforces this.
Win rate is the wrong metric. Target learning rate (>80% of experiments produce an actionable learning) over win rate. Teams optimizing win rate run safe, low-stakes tests.
NSM is not revenue. The North Star Metric measures value delivered to users. Revenue goes in guardrails. A product can increase revenue while losing product-market fit.
Loops beat funnels. Funnels require proportional reinvestment. Loops compound. /growth-loop-design tells you which you have and how to close the gap.
Growth skills support three motions:
| Motion | When it fits | Primary NSM focus |
|---|---|---|
| PLG (Product-Led Growth) | Self-deliverable value; short time-to-value; viable free tier | FVM rate, D30 retention |
| SLG (Sales-Led Growth) | Complex buyer; long sales cycle; high ACV | Pipeline velocity, champion activation |
| Community-Led (CLG) | Network effects; practitioners as advocates | Active contributors, time-to-first-answer |
Skills are motion-agnostic but prompt you to specify your motion — channel-to-motion mismatch is one of the most common experiment failure modes.
For motion framework definitions, PLG/SLG/MLG/CLG detail, lifecycle design, distribution norms, and paid acquisition, see DOMAIN.md.