The problem: a solo AI agency fails the same way every time — the coding agent starts writing code before anyone agrees on what is being built, and later "it runs" gets mistaken for "it's ready." What you need is not another toolbox; it's one connected pipeline where the agent must write down the spec, the plan and the tasks before touching code, and where each artifact can be checked by a human.
The solution in today's radar: one factory, one pipeline. GitHub Spec Kit installs slash commands into your AI coding agent that force a spec-driven workflow — constitution, specify, clarify, plan, tasks, implement, converge — with a human review point at every step.
One featured factory per day, per the format confirmed Nov 5, 2026. All links and the release details were verified from the repo's official pages on Nov 6, 2026; anything unverifiable is marked UNVERIFIED. The business-opportunity section is retired from this format. Prior daily reports were unavailable to this session, so repeat-coverage could not be checked; the v1.3.1 release (Nov 5) is a meaningful, verified update in any case.
/speckit-clarify step alone kills the biggest failure mode: the agent guessing what you meant. Every command has a defined human-review point (spec approval, plan approval, code/test evidence review).uv installer. No special hardware.| Agency OS stage | Spec Kit skill/module | Artifact produced | Which gate consumes it |
|---|---|---|---|
| PROBLEM_DEFINITION | /speckit.constitution | constitution.md (non-negotiable principles) | All later gates evaluate against it |
| SPECIFICATION | /speckit.specify + /speckit.clarify | feature spec + answered questions | Human/spec-approval gate before planning |
| ARCHITECTURE | /speckit.plan | technical plan (stack, interfaces, data model) | Human review of stack and constraints |
| QA (planning side) | /speckit.checklist + /speckit.analyze | "unit tests for English", consistency report | Gate before implementation |
| DEVELOPMENT | /speckit.tasks + /speckit.implement | dependency-ordered tasks, code diffs | Reviewer approves diffs and test evidence |
| MAINTENANCE | /speckit.converge | reconciliation of implemented reality vs spec | Post-implementation check |
| LEAD / DISCOVERY / PROPOSAL | No direct skill (honest gap) | — | — |
| SECURITY / RELEASE | No direct skill (honest gap) | — | — |
Spec Kit is the upstream half of a factory: problem definition through development and spec reconciliation. It complements gate_engine.py and never replaces it.
/speckit.implement output as production-ready.specify init in a client repo without reviewing what it scaffolds; keep experiments in a scratch repo.uv. Internet only for install. No API keys beyond your usual AI IDE subscription.uv tool install specify-cli --from git+https://github.com/github/spec-kit.git@v1.3.1specify init demo-speckit — or inside an existing scratch repo: specify init --here --integration claude (choose your AI IDE integration)./speckit.constitution with 3 of your Agency OS governance rules, then /speckit.specify for one tiny feature, then /speckit-clarify. Confirm the agent asks questions instead of assuming..specify/ folder with constitution.md, a spec doc, and a clarify transcript you can attach as evidence.advance_gate.ps1; evidence is marked VERIFIED only after a human read.