FAQ :: KNOWN UNKNOWNS

A tactical diagnostic engine for builders. You paste a GitHub repo URL, an AI chatlog, or a single word; we return a delusion score (0–100), a horizontal skill-tree of 3–5 project paths your seed could grow into, and on-demand artifacts (a 1-page PRD, a kill plan, a GitHub issue body, or a README badge) you can ship or paste straight into a repo.

No. scopecreeper.comis a B2B SaaS for enterprise consulting-spend management — it ingests invoices and tracks consulting engagements at Fortune 500 scale. Different audience, different problem, different price point. We're scopecreeper.ai: a creative diagnostic tool for indie builders and solo founders.

Two components, blended into one 0–100 score:
· Reality — how grounded the seed is (a real repo with active commits scores higher than a vague chat).
· Illusion — how ambitious it already is (how many features the chat claims, how broad the scope of the input is).
Low scores (corpse / sweetspot) mean shippable. High scores (abyss / delusion) mean: cool dream, but you'll be in the abyss in six weeks.

ChatGPT will happily say yes to your ambition and produce a 20-feature plan. Scope Creeper's job is the opposite — score the ambition, surface multiple paths, force you to pick one, and converge on a real artifact with a concrete stack and a 30-minute first cut. The LLM underneath is Llama 3.3 70B (free Workers AI, no Anthropic dep), but it's wrapped in a deliberately cynical product prompt.

Four terminal outputs you can converge on at any point in the tree:
· SHIPPABLE_V0.md — 1-page PRD with stack, V0 scope, acceptance criteria, and 4–6 paste-runnable shell commands.
· KILL.md — sunk-cost framing, specific dated cutoff signals, and a one-line eulogy.
· GH_ISSUE.md — title + labels + body, paste-ready for a GitHub issue.
· README_BADGE.svg— an embeddable score badge for any repo's README.

Yes. The ArtifactPanel has an EXPORT ▸ GH button. First click takes you through GitHub OAuth (public_repo scope only); then a repo picker. ISSUE artifacts open a real issue with the right labels; SHIPPABLE / KILL / BADGE create a new branch, commit the file underSCOPE_CREEPER/<kind>-<slug>, and open a PR.

Free tier: 5 scans per month, all branching, no artifacts, no share links. Pro: $9/month for unlimited scans, all artifact kinds, unlimited public share links, and 5 deep-audits per month. One-shot deep-audit: $5 per repo if you don't want a subscription.

An agentic Worker that pulls a public GitHub repo's tarball, walks the source, and runs grep heuristics for TODO density, dead/skipped tests, leaked secret patterns, stale dependencies, and debug-spam in non-test code. Every finding cites file + line. Narrative summary through Llama 3.3 70B, but evidence is real — not hallucinated. Capped at 200 files / 30MB / 30s wall.

By default yes — scan history is stored in your browser's localStorage; we never see the contents unless you explicitly hit the SHARE button (Pro-only). Shared threads write to a server-side KV keyed by a short slug; only people with the link can read them. Audit reports stay on your session.

Because Workers AI runs Llama 3.3 70B for free on Cloudflare's edge, and the prompt engineering matters more than the parameter count for this product. Zero Anthropic / OpenAI dependency by design — we can ship features without anyone's API key rotating mid-deploy.

Beers Labs LLC — built on top of an in-house diagnostic engine originally tuned for self-roasting our own delusional product plans. Open source-curious; pieces live at github.com/SipMyBeers/scopecreeper.