Nova Agent Workflows should improve over time without confusing existing users.
Use semantic versions:
MAJOR.MINOR.PATCH
MAJOR: breaking workflow or adapter changes.MINOR: new workflow, template, adapter, or substantial non-breaking capability.PATCH: wording fixes, safety clarifications, small template corrections.
Draft versions may use:
0.1.0-draft
0.3.x: Nova Serious Workflow v1.8 context contract, Goal/Rider task framing, structured risk approval, error budget, task-thread events, runtime harness verdicts, and agentic delivery chain
Included in v0.3:
nova-serious-workflownova-debug-review-gatenova-board-finisher- Codex skill adapter
- Claude Code skill adapter
- planning, task/checklist, work-log, and handoff templates
- synthetic examples only
- six-layer serious workflow: Obsidian Brain, Planning / Spec, GitHub Delivery, Verification / Risk Gate, Release / Publish, Sync Back / Learning
- planner/orchestrator/review overlay for app planning, feature planning, one-issue delivery, outside review, and clarification gates
- context contract before meaningful edits
- Goal/Rider task shape
- structured risk proposal before risky action
- allow/warn/review/block action verdicts
- repeated-failure retry budget
- compact task-thread events for handoff-heavy work
- agentic delivery chain from clarification to spec, issue plan, optional TDD, and architecture review
Potential v0.4 additions:
- HTML planning artifact workflow
- public-safe postmortem template
- graph/link hygiene workflow
- token/context budget workflow
- optional release checklist automation that only scans local files
- browser-game verification adapter
- Obsidian librarian adapter
- cross-agent handoff examples
- public test fixtures
Do not add account automation, deployment automation, secret handling, OS repair, or hooks that run by default.
Every upgrade should update:
VERSIONING.mdCHANGELOG.md- affected workflow files
- affected adapters
- examples or templates if behavior changed
- public scrub checklist result before release
- pre-publish release gate result before release
- cross-agent smoke review result before release
Every release candidate must pass the release gate in the same work session as the public action.
Required release-gate files:
safety/PUBLIC_EXPORT_SCRUB_CHECKLIST.md
safety/PRE_PUBLISH_RELEASE_GATE.md
safety/RISK_GATES.md
RELEASE_READINESS.md
The release gate expires after any edit to public files.
Do not create a GitHub repository, add a remote, push, publish, release, authenticate, or handle secrets as part of version alignment. Those actions require explicit approval for the exact action after the gate result is shown.
When private Nova workflows improve, decide whether the public pack should change too.
Update the public pack only when the improvement is reusable outside the private brain. Do not export private project memory, local runtime logs, account material, or machine-specific paths.
For each reusable upgrade, check:
workflows/*.md
adapters/codex/skills/*/SKILL.md
adapters/claude-code/skills/*/SKILL.md
adapters/claude-code/CLAUDE.md.example
templates/*.md
examples/*.md
safety/*.md
CHANGELOG.md
VERSIONING.md
If the change improves Nova's overall brain capability, also update the Nova Brain Evolution Dashboard in the private runtime.
Never publish, push, create a remote repository, or run account/auth/payment/secret actions as part of version alignment.
Public package versions and workflow revisions are tracked separately.
Current mapping:
package: 0.3.0-draft
nova-serious-workflow: v1.8
Use package semver for public release changes. Use workflow revision labels when the operating model changes.
Keep the core workflow files agent-agnostic. Put tool-specific details in adapters/.
If a workflow requires a feature only one agent supports, mark it as adapter-specific rather than changing the core workflow.