The repo-native record of intent and approvals for AI-assisted engineering — validated, attributable, readable by any agent.

WingFoil keeps your project's intent in git — tasks, specs, ADRs, decision logs, bugs — as typed documents with their own state machines. Every approval records who decided and why, in the commit. It works alongside the spec-driven tools and coding agents you already use, through a CLI and an MCP server.

npm install -g wingfoil
wingfoil init --template Scrum

flowchart LR
    T["wingfoil-templates<br/>packs"] --> W["wingfoil<br/>CLI + MCP"]
    W --> Y["your project"]
    B["wingfoil-benchmark<br/>measurements"] --> W
    U["wingfoil-ui<br/>local web console"] --> Y
    Y -. "feedback notes" .-> W

Repository	What it is	Latest
wingfoil	The core: CLI + MCP server — Memory, DNA, Directives, Workflows	v0.3.1 · npm 0.3.1 · docs · Listed on the M8ven Trust Index as a Verified Publisher.
wingfoil-templates	Versioned packs of templates, methodologies and directives that WingFoil consumes	base 2.0.1 · software 1.0.0 · methodology/scrum 1.0.0 · for WingFoil 0.3.0–0.3.1, adopted by hand
wingfoil-ui	Local web console for WingFoil-governed projects: read a project's Memory, DNA, directives and team in the browser	v0.2.0 · npm 0.2.0, npx wingfoil-ui · works with WingFoil 0.2.x
wingfoil-benchmark	Measures where WingFoil makes a difference, against plain agents and other tools	v0.2 results (preliminary; Spec Kit and OpenSpec arms) · Codex baseline, across agents (preliminary) · Codex execution 13fda3099afa/1 (with an erratum; no rerun, the erratum's note stands)

More: how the repositories relate.

• 📚 For users — packages, guides, releases and benchmark results
• 🤖 For AI agents — MCP server and the files written for agents
• 🔁 Feed back into WingFoil — how a project using WingFoil reports defects and requests, and learns which release answered them

• 💬 Discussions — questions and ideas
• 🐞 Issues — bugs and proposals
• 🤝 How to contribute — through Memory, not (only) code

• Every repository listed above is governed with WingFoil itself: its Memory is public under docs/.
• The benchmark is built and run by WingFoil's maintainer (validity threat T1): its rules are fixed before any run, losses are published like wins, and every run's record, cost and transcript are public.
• Security reports: see SECURITY.md.
• All repositories are MIT licensed.