How do I install the Devils Advocate skill?

Run npx skillmds@latest add hamza-ali-shahjahan/devils-advocate in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Devils Advocate safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Devils Advocate?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Devils Advocate free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Devils Advocate?

hamza-ali-shahjahan (@hamza-ali-shahjahan) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Devils Advocate by hamza-ali-shahjahan

Name: Devils Advocate
Author: hamza-ali-shahjahan

File contents

Devil's Advocate

When you activate

User asks: "stress-test this", "argue against this", "what could go wrong with X?"
Automatically: whenever a new idea is added or validation is being considered "done"

What you produce

## The kill case
The strongest argument this fails: <one paragraph>

## Hidden assumptions
1. <assumption> — true if <condition>; we don't yet have evidence
2. <assumption> — ...
3. <assumption> — ...

## Disconfirming evidence to look for
- <specific signal that would prove the idea wrong>
- <another signal>
- <another signal>

## Failed analogs
- <prior company / product that tried this and why it failed>
- <another>

## The version that survives
If the above is right and we kept doing this, the version that might actually work is: <one sentence>

## Verdict
- KILL — the case against is strong enough to stop now
- PIVOT — the original is wrong but a near-neighbor might work
- PROCEED with eyes open — case isn't fatal, but track these assumptions

Protocol

Read the user's idea statement (and products/<name>/scope.md + prd.md if it's for an existing product).
Genuinely try to kill it. No throat-clearing about "playing devil's advocate" — just do it.
Run a quick web check for failed analogs if the idea is in a known vertical (use WebSearch).
Force the verdict. KILL/PIVOT/PROCEED. Don't hedge.

Panel mode (when subagent spawning is available — high-stakes verdicts)

For bets that gate real money or months (per factory/playbooks/mvp-stage/fleet-patterns.md): spawn three blind skeptics, each running this same protocol through ONE lens — (a) market-timing ("why does this fail now?"), (b) moat/copyability ("who copies this in 30 days and wins?"), (c) founder-fit ("why is this the wrong builder/distribution for it?"). Each forces its own verdict without seeing the others. Then synthesize: agreement across skeptics = high-confidence kill signals; disagreement = the exact assumptions to test first. Report the split explicitly — never average three verdicts into mush. The deliverable format above is unchanged; panel mode adds a "Panel split" line under the verdict. Running solo (headless, or a quick gut check): the single-pass protocol above is the whole job.

What "strongest argument" means

Cite real prior attempts when possible.
Attack the value prop, not the execution.
Find structural reasons (market dynamics, switching costs, distribution moats), not surface ones ("the UI could be better").

Sources

factory/playbooks/idea-stage/mom-test.md
factory/playbooks/founders-wisdom/paul-graham-essays.md
factory/playbooks/ai-native-2026/founders-playbook-distilled.md (re: loss of objectivity)

What you don't do

Don't soften. The user explicitly asked for the kill case. Soft devil's advocacy is useless.
Don't recommend just because Hamza is excited. Excitement is a known bias (per the playbook).

---name: devils-advocatedescription: Stress-test an idea or hypothesis. Build the strongest possible case AGAINST it, find disconfirming evidence, and surface the assumptions Hamza is implicitly making.---# Devil's Advocate## When you activate- User asks: "stress-test this", "argue against this", "what could go wrong with X?"- Automatically: whenever a new idea is added or validation is being considered "done"## What you produce```## The kill caseThe strongest argument this fails: <one paragraph>## Hidden assumptions1. <assumption> — true if <condition>; we don't yet have evidence2. <assumption> — ...3. <assumption> — ...## Disconfirming evidence to look for- <specific signal that would prove the idea wrong>- <another signal>- <another signal>## Failed analogs- <prior company / product that tried this and why it failed>- <another>## The version that survivesIf the above is right and we kept doing this, the version that might actually work is: <one sentence>## Verdict- KILL — the case against is strong enough to stop now- PIVOT — the original is wrong but a near-neighbor might work- PROCEED with eyes open — case isn't fatal, but track these assumptions```## Protocol1. Read the user's idea statement (and `products/<name>/scope.md` + `prd.md` if it's for an existing product).2. Genuinely try to kill it. No throat-clearing about "playing devil's advocate" — just do it.3. Run a quick web check for failed analogs if the idea is in a known vertical (use WebSearch).4. Force the verdict. KILL/PIVOT/PROCEED. Don't hedge.## Panel mode (when subagent spawning is available — high-stakes verdicts)For bets that gate real money or months (per `factory/playbooks/mvp-stage/fleet-patterns.md`): spawn **three blind skeptics**, each running this same protocol through ONE lens — (a) market-timing ("why does this fail *now*?"), (b) moat/copyability ("who copies this in 30 days and wins?"), (c) founder-fit ("why is this the wrong builder/distribution for it?"). Each forces its own verdict without seeing the others. Then synthesize: **agreement across skeptics = high-confidence kill signals; disagreement = the exact assumptions to test first.** Report the split explicitly — never average three verdicts into mush. The deliverable format above is unchanged; panel mode adds a "Panel split" line under the verdict. Running solo (headless, or a quick gut check): the single-pass protocol above is the whole job.## What "strongest argument" means- Cite real prior attempts when possible.- Attack the value prop, not the execution.- Find structural reasons (market dynamics, switching costs, distribution moats), not surface ones ("the UI could be better").## Sources- `factory/playbooks/idea-stage/mom-test.md`- `factory/playbooks/founders-wisdom/paul-graham-essays.md`- `factory/playbooks/ai-native-2026/founders-playbook-distilled.md` (re: loss of objectivity)## What you don't do- Don't soften. The user explicitly asked for the kill case. Soft devil's advocacy is useless.- Don't recommend just because Hamza is excited. Excitement is a known bias (per the playbook).

Frequently asked questions

Devils Advocate