nova-specSpec-Driven Development for Claude Code

From a ticket to a merged PR in explicit steps, with architectural memory that doesn't decay.

Get started

View on GitHub

Disciplined ticket → PR flow

Nine /nova-* slash commands enforce sequence, file structure, and naming. Skip a step and the next one stops you.

Atomic markdown memory

Decisions, gotchas, and services in one-fact-per-file markdown. No DSL, no daemon, no database.

Jira-first, multi-forge

Native Jira integration as a skill. Auto-detects GitHub (gh) and GitLab (glab) from your git remote.

Auto-sync that doesn't clobber

Hash-compares every framework file on every Claude Code session start. Your local edits are always preserved.

At a glance

/nova-start  →  /nova-spec  →  /nova-plan  →  /nova-build  →  /nova-review  →  /nova-wrap

Each command has a guardrail. Each guardrail is deterministic (bash + file existence, never LLM judgment). Skip a step and the next one stops you. Edit any framework file in place — sync hash-compares and never overwrites your changes.

Where to go next

First time? → Getting started
Want to know what each command does? → Flow overview
Want to customize the PR template, review checklist, or commit format? → Customization overview
Want to look up a config key or CLI subcommand? → Reference
Hit an error? → Troubleshooting

Principles


Form, not substance	Guardrails check sequence and file shape, not quality. Quality is your job.
Plain markdown	No DSLs, no schema, no build step. If you can `grep` it, it's source.
No hidden state	Everything in git. No daemon, no database, no cloud component.
Atomic memory	One fact = one file. Filename is the index. Supersede explicitly.
You can always escape	Skip a step, edit a command, bypass the framework — it never blocks shipping.

The full philosophy lives in PHILOSOPHY.md of the repo and is the antibody to scope creep.