ZeroGuess

Stop guessing. Start contracting.

ZeroGuess finds every implementation fork, scores them by impact and reversibility, auto-resolves what your project already decides, asks about the rest, and writes a machine-readable .zeroguess contract that enforces decisions during implementation.

Not a chatty Q&A session. A 2-5 minute interrogation that produces an enforceable contract.

The Problem

Every clarification tool — including good ones — produces prose summaries that get forgotten the moment implementation starts. The agent asks 4 great questions, gets answers, writes a paragraph, then silently makes 20 more decisions on its own.

ZeroGuess fixes the full pipeline: clarify → score → contract → enforce.

How It Works

Request
  ↓
Silent Fork Analysis (internal)
  - Score each fork: impact × reversibility × context available
  - Auto-resolve: project already has Tailwind? Lock it, don't ask.
  - Kill: low-impact, easily reversible? Skip.
  ↓
Declare Auto-Resolved (show what you already know)
  "Stack: Next.js, Prisma, Tailwind ✓ — correct me if wrong"
  ↓
Ask Remaining Forks (max 5, one at a time)
  Each with Surprise Me option
  ↓
Write .zeroguess Contract (machine-readable YAML)
  ↓
Implementation: agent checks contract before every decision
  - In contract → use locked value
  - In open_questions → STOP and ask
  - Not mentioned + high impact → STOP and ask

The .zeroguess Contract

# .zeroguess — Decision Contract
project: crypto-dashboard
created: 2026-03-15

decisions:
  platform:
    value: web
    source: user
    impact: high
    reversibility: hard
    locked: true

  database:
    value: PostgreSQL via Prisma
    source: auto-resolved   # found prisma/schema.prisma
    impact: high
    reversibility: hard
    locked: true

open_questions:
  - "Caching strategy"
  - "Error monitoring"

Agents read this file. Locked decisions are respected. Open questions trigger a stop-and-ask. No more silent assumptions.

Fork Scoring

Not fake math. Categorical heuristics:

Dimension	High	Medium	Low
Impact	Changes DB + API + frontend	Changes one layer	Cosmetic
Reversibility	Migration needed	Refactor needed	Swap anytime
Context	Nothing hints at answer	Partial clues	Already decided

• HIGH impact + HARD to reverse + NO context → must ask
• HIGH impact + context available → auto-resolve (declare, don't ask)
• LOW impact + easy to reverse → skip

What's Different

Feature	Typical clarifier	ZeroGuess
Output	Prose summary (forgotten)	.zeroguess contract (enforced)
Project awareness	Sometimes reads files	Scores forks against project context
Auto-resolve	No	Yes — declares known answers, skips questions
Implementation enforcement	None	Contract checked during coding
Cross-session memory	None	.zeroguess persists in repo
Surprise Me	No	Every question
Question limit	None	Hard cap at 5

Install

# For GSD / pi users
mkdir -p ~/.agents/skills/zeroguess

# For Claude Code users  
mkdir -p ~/.claude/skills/zeroguess

Copy SKILL.md into the directory. Restart your agent.

Usage

/zeroguess        # standalone
/gsd-zeroguess    # with GSD

The .zeroguess file is committed to your repo. Any agent that reads it will respect your decisions.

License

MIT

Credits

Built by @toismfer

⭐ Star if it saves you from silent assumptions