Business Idea Filter (BIF)

A Claude Code skill suite that pressure-tests business ideas against your own founder profile and criteria — not a generic rubric. You brain-dump an idea; it interviews you where needed, runs an adversarial office-hours stress test, fans out competitor and SEO research, scores the idea against a weighted rubric tuned to your ambition tier, and ends with the one thing that matters most: the single riskiest assumption, and the cheapest experiment to test it this week.

Scope: BIF evaluates the business opportunity — demand, distribution, defensibility, founder-market fit, economics, durability. It does not plan architecture, tech stack, or implementation. If a session drifts into "how to build it," the advisor names the boundary and redirects. Dev planning is a different tool.

Why it's different

The same idea is a great bet for one founder and a terrible one for another. BIF keeps your rubric and ambition tier (VC / bootstrapped / services) in a persistent profile, and every evaluation is scored against that. A bootstrapper sees "some competition is a green flag"; a VC sees "crowded unless there's a sharp wedge." Same idea, honestly different verdicts.

Install

/plugin marketplace add njcameron/bif         # or your fork's owner/repo
/plugin install business-idea-filter@bif
/reload-plugins

The skills

Skill	Role
profiling-founders	One-time onboarding interview → persistent founder profile (your rubric lives here).
validating-ideas	The lead advisor. Orchestrates the whole evaluation and produces the verdict.
researching-competitors	Web subagent: prior art, the feature/product/company verdict, white space.
analyzing-seo-landscape	Routes to the claude-seo toolkit for keyword opportunity + content angle.

The adversarial office-hours stress test runs as a phase inside validating-ideas.

First run

1. Set up your profile — say "set up my idea board" or invoke profiling-founders. It asks five spoken-style questions (track record, constraints, goal, unfair advantages, ambition tier), optionally ingests your website, and writes a profile you can hand-edit any time.
2. Validate an idea — say "is this worth building: …" or invoke validating-ideas, then brain-dump. You'll get a tier-framed verdict, a per-lens scorecard, the riskiest assumption, and a concrete experiment to run this week.

Where your data lives

Your founder profile and one folder per idea live under ~/idea-board by default. Set the IDEA_BOARD_ROOT environment variable to put them elsewhere (e.g. inside an Obsidian vault):

export IDEA_BOARD_ROOT="$HOME/Obsidian/MyVault/idea-board"

Each idea is a folder, not a file: the synthesized dossier.md, a machine-readable scorecard.json, your verbatim brain-dump, and a quarantined evidence/ tree of everything the research stages captured — so a verdict stays auditable months later. This data is yours and is never part of the plugin repo.

Optional dependencies

BIF works without these, degrading gracefully — but they make it better:

• Firecrawl — for ingesting your website during onboarding. Connect a Firecrawl MCP server, or set FIRECRAWL_API_KEY. (Claude Code's built-in WebFetch is used automatically if available, so this is rarely needed.)

• claude-seo — for live keyword/SERP data in the SEO lens. Install the agricidaniel-seo marketplace and the claude-seo plugin. Without it, the SEO lens gives a qualitative read only.

  If you'd rather make claude-seo a hard, auto-installed dependency, add it to plugins/business-idea-filter/.claude-plugin/plugin.json under a dependencies array and allowlist its marketplace in .claude-plugin/marketplace.json via allowCrossMarketplaceDependenciesOn. It's left optional by default so BIF doesn't force a DataForSEO setup on everyone.

Security model

Anything fetched from the web (competitor pages, your own marketing site) is treated as data, never instructions — a page that says "ignore previous instructions" is quoted to you, not obeyed. Research runs in subagents, isolated from anything that writes your profile.

Attribution

The office-hours stress-test phase is adapted from the office-hours skill in gstack by Garry Tan, used under the MIT License. It is a derivative: the gstack runtime plumbing and the promotional coda were removed, and it was refit to run as a dossier-writing phase. See plugins/business-idea-filter/LICENSE for full notices.

License

MIT — see plugins/business-idea-filter/LICENSE.