Engineering decisions that know when they're stale.
Frame problems. Compare options fairly. Record decisions as contracts. Know when to revisit.
Supports: Claude Code, Cursor, Gemini CLI, Codex CLI, Codex App, Air
curl -fsSL https://raw.githubusercontent.com/m0n0x41d/quint-code/main/install.sh | bashThen in your project, run init with your tool's flag:
# Claude Code (default if no flag)
quint-code init
# Cursor
quint-code init --cursor
# Gemini CLI
quint-code init --gemini
# Codex CLI / Codex App
quint-code init --codex
# JetBrains Air
quint-code init --air
# All tools at once
quint-code init --allThe binary is the same — only the MCP config location differs:
| Tool | MCP Config | Commands | Skill |
|---|---|---|---|
| Claude Code | .mcp.json (project root) |
~/.claude/commands/ |
~/.claude/skills/q-reason/ |
| Cursor | .cursor/mcp.json |
~/.cursor/commands/ |
~/.cursor/skills/q-reason/ |
| Gemini CLI | ~/.gemini/settings.json |
~/.gemini/commands/ |
— |
| Codex CLI | .codex/config.toml |
~/.codex/prompts/ |
~/.agents/skills/q-reason/ |
| Air | .codex/config.toml |
project skills/ |
project skills/q-reason/ |
Important for Cursor: After init, open Cursor Settings → MCP → find quint-code → enable the toggle. Cursor adds MCP servers as disabled by default.
Note: Cursor also picks up Claude Code commands from ~/.claude/commands/ — so slash commands work even without --cursor. But MCP config (.cursor/mcp.json) must be set up for the tools to connect.
Existing project? Run /q-onboard after init — the agent scans your codebase for existing decisions worth capturing.
First time? Ask the agent to explain how it works:
/q-reason explain how to work with quint-code effectively — what commands exist, when to use each one, and what's the recommended workflow
The agent has full knowledge of all Quint tools and will walk you through them in context of your project.
Describe your problem. The agent frames it, generates alternatives, compares them fairly, and records the decision — all in one command. It auto-selects the right depth.
/q-frame → /q-char → /q-explore → /q-compare → /q-decide
what's what genuinely fair engineering
broken? matters? different comparison contract
options
The agent captures decisions automatically when it notices them in conversation. No rationale — no record. Conflicts with active decisions are flagged. Auto-expires in 90 days.
/q-status shows what's expired and what needs attention. /q-refresh manages the lifecycle of ALL artifact types — waive, reopen, supersede, or deprecate.
- Decisions are live — they have computed trust scores (R_eff) that degrade as evidence ages. An expired benchmark drops the whole score.
- Comparison is honest — parity enforced, dimensions cross-checked, asymmetric scoring warned. Anti-Goodhart: tag dimensions as "observation" to prevent optimizing the wrong metric.
- Memory across sessions — when you frame a problem, the tool surfaces related past decisions. When you explore, it checks for similar variants.
- The loop closes — failed measurements suggest reopening. Evidence decay triggers review. Periodic refresh prompts ensure nothing goes stale silently.
- Decisions are contracts — FPF E.9 format: Problem Frame, Decision (invariants + DO/DON'T), Rationale, Consequences. A new engineer reads it 6 months later and gets everything.
| Tool | What it does |
|---|---|
quint_note |
Micro-decisions with validation + auto-expiry |
quint_problem |
Frame problems, define comparison dimensions with roles |
quint_solution |
Explore variants with diversity check, compare with parity |
quint_decision |
FPF E.9 decision contract, impact measurement, evidence |
quint_refresh |
Lifecycle management for all artifacts |
quint_query |
Search, status dashboard, file-to-decision lookup |
FPF by Anatoly Levenchuk — a rigorous, transdisciplinary architecture for thinking.
/q-reason gives your AI agent an FPF-native operating system for engineering decisions: problem framing before solutions, characterization before comparison, parity enforcement, evidence with congruence penalties, weakest-link assurance, and the lemniscate cycle that closes itself when evidence ages or measurements fail.
quint-code fpf search gives you access to 4243 indexed sections from the FPF specification — the agent can look up any concept on demand.
See the documentation for detailed guides on decision modes, the DRR format, computed features, and lifecycle management.
- Go 1.24+ (for building from source)
- Any MCP-capable AI tool
MIT