← ClaudeAtlas

writing-adrslisted

Use when authoring an Architecture Decision Record (ADR) — a short document that captures one significant, hard-to-reverse technical or architectural decision, its context, and its consequences. Triggered by the sdlc ADR pipeline (the knowledge-engineer agent behind /sdlc:docs seed adr and /sdlc:docs distill) when generating an ADR from a story, and by anyone hand-authoring an ADR under docs/adr/ today. Covers why ADRs are kept short and inverted-pyramid, the required sections (Title, Status, Decision, Context, Alternatives Considered, Consequences), the proposed→accepted→superseded (or →rejected) status lifecycle and the never-edit-only-supersede immutability rule, the NNNN-decision-slug.md filename convention, and the ADR frontmatter fields (status, agents, source-stories) the pipeline reads to route a generated ADR into docs/adr/index.md.
whimzyLive/nightshift-ai · ★ 3 · AI & Automation · score 71
Install: claude install-skill whimzyLive/nightshift-ai
# writing-adrs — Architecture Decision Records **Core principle:** an ADR records ONE decision — not a design, not a feature, not a whole subsystem. If you find yourself wanting to describe several decisions in one file, that's a signal to split it into several ADRs. **Announce at start:** "I'm using the writing-adrs skill to author this ADR." ## When to Use - Whenever a decision is architecturally significant: it affects structure, non-functional characteristics (performance, security, cost), external dependencies, interfaces between systems, or is expensive to reverse later. - When the sdlc ADR pipeline (`/sdlc:docs seed adr` / `/sdlc:docs distill` + `knowledge-engineer` agent) generates an ADR from a Jira story or a completed piece of work — in a repo where that pipeline is adopted. - Not for routine implementation choices that any competent engineer would make the same way and that cost nothing to change later — those don't need a permanent record. ## Why ADRs Look the Way They Do This isn't an arbitrary template — every rule below exists to solve a specific failure mode. Understanding the "why" matters more than following the shape mechanically, because a generated ADR that hits every section heading but skips the reasoning behind them is not actually useful to the future reader it's written for. - **One decision per record, not one document per subsystem.** A record that bundles several decisions together can't be individually superseded — the moment