bmad-forge-idea

Pressure-tests ideas through persona-driven interrogation until they harden, clarify, or die.

Updated Sep 3, 2026
One-click install
npx skills add https://github.com/watchthelight/shatterfish --skill bmad-forge-idea-watchthelight
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: bmad-forge-idea
Source: https://github.com/watchthelight/shatterfish/tree/main/.claude/skills/bmad-forge-idea
Command: npx skills add https://github.com/watchthelight/shatterfish --skill bmad-forge-idea-watchthelight

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Half-formed ideas often carry unchecked assumptions and unresolved decisions that become expensive problems later. This Skill stress-tests an idea in conversation, while changing your mind is still cheap, so you can act on it with conviction or reject it early. ## Core Features & Use Cases - Persona-Driven Interrogation: Cross-examines your idea using installed BMad agents, custom personas, and saved party groups, plus a freshly generated outside voice such as a competitor, buyer, or critic. - Attack/Defend Modes: Say "attack this", "defend this", or "switch roles" at any time to change how the idea is argued; attack mode never agrees with the idea. - Persistent Session State: Records every decision, assumption, crack, kill, and locked idea to an on-disk memlog, so sessions survive interruption and can be resumed later. - Structured Outcomes: Ends as Hardened (distills a short forged-idea.md for downstream planning skills), Killed (with cause of death), or Clearer, and always renders a self-contained forge-report.html with a wax-seal-style outcome stamp. - Use Case: You have a rough product feature idea. Invoke the forge, and it questions your assumptions one branch at a time, brings in a skeptical buyer persona and a finance reviewer, locks the decisions that survive, and hands off a hardened idea file to your planning workflow. ## Quick Start Ask the assistant to forge or pressure-test your idea, for example by saying "forge this idea: a subscription tier for our open-source tool". ## Quick Start Tell the assistant to pressure-test your idea, such as "stress-test my idea for a mobile habit-tracking app".

Frequently Asked Questions about bmad-forge-idea

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I pressure-test a product idea with AI personas?

Invoke the forge by saying "forge an idea" or "pressure-test this idea" and describe your idea. The skill interrogates it one question at a time using installed BMad agents, custom personas, and a generated outside voice like a competitor or buyer.

What is the difference between attack mode and defend mode?

Attack mode never agrees with the idea and hunts for contradictions, weak assumptions, and failure cases. Defend mode argues for the strongest version of the idea. Say "attack this", "defend this", or "switch roles" at any time to change modes.

Can I resume an interrupted idea-forging session?

Yes. Session state is written to an on-disk memlog after the goal is known. On activation the skill scans for memlogs whose status is not complete and offers to resume one, rebuilding state from the recorded decisions and assumptions.

What outputs does the idea forge produce?

Every session renders a self-contained forge-report.html with an outcome seal marked HARDENED, KILLED, or CLARIFIED. Hardened ideas also produce a short forged-idea.md containing only the locked decisions, rejected options, and reasons for downstream planning skills.

Does the forge work without BMad party-mode installed?

Yes. The resolve_personas.py script falls back to reading user override TOMLs directly when bmad-party-mode is not found, and if no persona pool is available the skill generates the interrogating voices itself.

When should I not use persona-based idea forging?

Avoid it when you need a deliverable document rather than better thinking, since the skill's goal is interrogation, not artifact production. It also does not steer toward build decisions, so use planning skills like bmad-spec or bmad-prd afterward for execution.