Skip to content

💫 Toolkit to help you get started with SDD or any other process!

docs: document spec persistence models (#2856)

Wondr0b82a1ddf1d0
diff --git a/docs/concepts/sdd.md b/docs/concepts/sdd.md
index 7f59feb51..c5b6d5542 100644
--- a/docs/concepts/sdd.md
+++ b/docs/concepts/sdd.md
@@ -11,6 +11,11 @@ Spec-Driven Development is a structured process that emphasizes:
- **Multi-step refinement** rather than one-shot code generation from prompts
- **Heavy reliance** on advanced AI model capabilities for specification interpretation
+Spec Kit does not prescribe how teams preserve or mutate `spec.md`, `plan.md`,
+and `tasks.md` after requirements change. See
+[Spec Persistence Models](spec-persistence.md) for three common ways to manage
+those artifacts over time.
+
## Development Phases
| Phase | Focus | Key Activities |
diff --git a/docs/concepts/spec-persistence.md b/docs/concepts/spec-persistence.md
new file mode 100644
index 000000000..dcef44111
--- /dev/null
+++ b/docs/concepts/spec-persistence.md
@@ -0,0 +1,107 @@
+# Spec Persistence Models
+
+Spec Kit intentionally leaves teams in control of what happens to `spec.md`,
+`plan.md`, and `tasks.md` after requirements change. The toolkit gives you a
+repeatable workflow, but it does not force one artifact maintenance strategy.
+
+This page names three common models so teams can make that choice explicit.
+None is the default, and none is required by Spec Kit.
+
+## Two Separate Questions
+
+Spec-driven development has a temporal question: how long should the
+specification matter? One
+[overview of SDD tooling](https://martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html)
+frames that lifecycle in three levels:
+
+- **Spec-first**: write a spec before coding, then allow it to be discarded.
+- **Spec-anchored**: keep the spec after implementation and use it for future
+ changes.
+- **Spec-as-source**: treat the spec as the only human-edited source and
+ regenerate implementation artifacts from it.
+
+Spec Kit also exposes a second question: what happens to the artifact set when
+requirements change? The models below describe that mutation strategy.
+
+## Flow-Back Spec
+
+Use flow-back when `spec.md`, `plan.md`, `tasks.md`, and the implementation are
+all allowed to inform each other.
+
+In this model, edits can begin in any artifact. A developer might update
+`tasks.md` during implementation, revise `plan.md` after a technical discovery,
+or adjust `spec.md` after a product clarification. The team then reconciles the
+artifact set manually so the final project history still makes sense.
+
+Flow-back works well when:
+
+- the team is small enough to notice and reconcile drift quickly
+- implementation discoveries are expected to reshape the original plan
+- speed matters more than preserving each intermediate decision as immutable
+ history
+
+The main risk is silent divergence. If the team changes lower-level artifacts
+without reflecting the decision back into `spec.md`, future contributors may
+not know which artifact to trust.
+
+## Flow-Forward Spec
+
+Use flow-forward when each feature directory should remain a historical record.
+
+In this model, completed artifacts are treated as immutable. When requirements
+change, the team creates a new feature directory instead of mutating the
+existing `spec.md`, `plan.md`, or `tasks.md`. The older directory remains useful
+for audit, comparison, or explaining how the project reached its current state.
+
+Flow-forward works well when:
+
+- auditability and traceability matter
+- features are well-scoped and rarely revisited in place
+- the team wants a clear sequence of requirement changes over time
+
+The main tradeoff is duplication. Related decisions can be spread across
+multiple feature directories, so teams need naming, linking, or review habits
+that make the lineage easy to follow.
+
+## Living Spec
+
+Use living spec when `spec.md` is the contract and the other artifacts are
+derived from it.
+
+In this model, teams update `spec.md` first and then regenerate or revise
+`plan.md` and `tasks.md` from that source. The plan and task list are still
+valuable, but they are treated as disposable derivations rather than permanent
+sources of truth.
+
+Living spec works well when:
+
+- the product contract is stable enough to own the workflow
+- the team is comfortable regenerating derived artifacts after spec changes
+- consistency between requirements and implementation matters more than keeping
+ every intermediate plan intact
+
+The main risk is losing useful implementation rationale if derived artifacts are
+discarded without preserving important decisions elsewhere.
+
+## Choosing a Model
+
+The model is a team convention, not a CLI setting. A project can even use
+different models in different areas, as long as contributors know which one
+applies.
+
+| Model | Mutation rule | Best fit | Watch out for |
+|---|---|---|---|
+| Flow-back spec | Edit any artifact, then reconcile | Fast iteration and close collaboration | Silent drift between artifacts |
+| Flow-forward spec | Create a new feature directory for new requirements | Audit trails and historical clarity | Duplicate or fragmented context |
+| Living spec | Edit `spec.md`; regenerate derived artifacts | Spec as contract | Lost rationale in regenerated files |
+
+If your team has not chosen a model yet, start by answering two questions:
+
+1. Should completed feature directories be historical records or editable work
+ areas?
+2. Is `spec.md` the single source of truth, or are `plan.md` and `tasks.md`
+ allowed to become co-equal sources?
+
+Once those answers are clear, document the convention in your project
+constitution or team onboarding notes so future contributors know how to handle
+changes.
diff --git a/docs/toc.yml b/docs/toc.yml
index ce91339a1..e5fb07fbb 100644
--- a/docs/toc.yml
+++ b/docs/toc.yml
@@ -41,6 +41,8 @@
items:
- name: What is SDD?
href: concepts/sdd.md
+ - name: Spec Persistence Models
+ href: concepts/spec-persistence.md
# Development workflows
- name: Development