// the find
github/spec-kit
💫 Toolkit to help you get started with SDD or any other process!
Spec Kit is GitHub's toolkit for running AI coding agents through structured processes — spec-driven development (specify → plan → tasks → implement → converge), plus opt-in bug-fixing and idea-assessment workflows. It's for teams who want their AI agent sessions to produce a paper trail (specs, plans, verdicts) instead of ad-hoc prompting, across tools like Copilot, Claude Code, etc.
The extension/preset/bundle split keeps the core SDD flow small while letting you opt into bug-fixing or idea-assessment only when you need them, instead of one monolithic workflow. The bug-fixing process explicitly separates diagnosis, repair, and verification into distinct skills with a recorded verdict (verified/partial/failed), which stops an agent from marking its own fix successful without checking the original symptom. It's agent-agnostic by design — same processes, different integration key per coding agent — rather than being hard-wired to one vendor's tool.
The README is almost entirely a funnel to the docs site; you can't tell what a `/speckit-plan` or `/speckit-converge` invocation actually does to your repo without digging through `.specify/` templates and extension command files yourself. It's a process scaffold, not a correctness tool — all the actual work (and all the actual risk of bad code) still comes from the underlying agent, so a disciplined-looking pipeline can still produce confidently wrong specs and plans. Requiring Python 3.11+ and uv just to template prompts and manage markdown skill files is a surprising amount of tooling for what is fundamentally string generation. Cross-agent support means invocation syntax differs per integration, so the one-liner examples in the README don't directly transfer to whichever agent you're actually using.