Skip to content
Back to skills

Validate

ASecurity

Run a full validation pass on an idea — sharpening, devil's advocate, market sizing, competitor mapping, customer discovery plan.

  • 9 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 5, 2026
ai-agentsgo

Security analysis

A100/100

Scanned September 5, 2026

npx -y skills add hamza-ali-shahjahan/hamzaish --skill validate --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Validate?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Validate
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/hamza-ali-shahjahan-validate-hamzaish/badge)](https://www.skillsdirectory.com/skills/hamza-ali-shahjahan-validate-hamzaish)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: validate
description: Run a full validation pass on an idea — sharpening, devil's advocate, market sizing, competitor mapping, customer discovery plan.
---

# /validate

An opt-in rail, not a gate. Use `/validate` when a bet is expensive, slow, or hard to undo — paid acquisition, a big build, a pivot. For cheap, reversible ideas, just build; the ship is the validation.

Usage: `/validate <idea or product name>`

## What this does
Runs the Idea-stage agents in sequence (or parallel where independent):
1. **`problem-sharpener`** — turn the idea into a testable hypothesis
2. **`devils-advocate`** — build the strongest case against
3. **`market-researcher`** — TAM/SAM/SOM + trends
4. **`competitor-mapper`** — landscape by tier
5. **`customer-discovery`** — interview plan + outreach templates

Outputs go to `products/<slug>/` (creating the folder if new).

## What you do as the assistant when this is invoked
1. Resolve `<idea>` to a product slug (existing or new).
2. If new product: create `products/<slug>/` with minimal structure, log in `decisions/0001-validation-start.md`.
3. Run the 5 agents — **fleet mode when subagent spawning is available** (see `factory/playbooks/mvp-stage/fleet-patterns.md`):
   - **Fan out blind:** problem-sharpener, devils-advocate, market-researcher, competitor-mapper as 4 concurrent subagents, each getting the raw idea + its own SKILL.md only — none sees another's output (independent verdicts are the point; devils-advocate attacks the raw idea, which is the harder, more honest target). Spawn each on its `model_tier`.
   - **Verify adversarially:** before synthesis, spawn one refuter against the kill case ("is this actually fatal, or survivable?") and one against the strongest FOR evidence ("is this signal or founder hope?").
   - **Synthesize:** customer-discovery runs on the (surviving) sharpened hypothesis; then produce the snapshot, reporting where the blind agents *agreed* (high confidence) and *disagreed* (the open questions — never averaged away).
   - **Serial fallback (no spawning):** run in sequence as before — 1, then 2 on 1's output, 3+4, then 5 on 1+2. Same deliverables either way.
4. After all done: produce a 1-page "validation snapshot" summarizing GO / PAUSE / KILL with reasoning — including the agreement/disagreement map when fleet mode ran.

## Output location
- `products/<slug>/validation/snapshot.md` — the summary
- `products/<slug>/validation/hypothesis.md` — sharpened problem
- `products/<slug>/validation/devils-advocate.md` — kill case
- `products/<slug>/validation/market.md` — sizing + trends
- `products/<slug>/validation/competitors.md` — landscape
- `products/<slug>/interviews/discovery-plan.md` — the plan

## When NOT to invoke
- If the idea hasn't been articulated in writing yet — ask user to write 2 sentences first.
- If user just wants a quick gut check — that's `idea-generator` or `devils-advocate` directly, not full validation.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.