Arcadi4 Writing ADRs
Overview
Write Architecture Decision Records as lightweight decision snapshots, not mini design docs.
Core principle: Record one meaningful decision, why it was chosen, what tradeoffs came with it, and what should trigger reconsideration later.
This skill adapts to project scale — solo, hobby group, small team, or business — instead of assuming a committee.
When to Use
Use this skill when:
- A design or planning discussion has produced a real decision
- The user asks to capture, document, or formalize a design choice
- A decision will likely be questioned later
- A choice affects structure, quality attributes, tooling, interfaces, deployment shape, or a hard-to-reverse workflow
- You need a durable record of why one option beat another
Use it after design discussion, not before. If the decision is still unsettled, keep designing.
Do not use this skill for:
- Implementation notes or rollout checklists
- Obvious local edits with no lasting tradeoff
- Broad solution designs containing multiple independent decisions
- Meeting transcripts or brainstorming dumps
Adaptive Metadata by Project Scale
The ADR header is not one-size-fits-all. Choose metadata based on project context.
Determine the scale BEFORE writing the ADR. If uncertain, ask the human partner.
| Scale | Metadata | Key rule | Guide | Template |
|---|---|---|---|---|---|
| Solo | Status, Date | Ask the human first. Do not auto-apply. | scale-guides/solo-projects.md | templates/solo-adr-template.md |
| Hobby group | Status, Date, Involved | Lightweight. No governance ceremony. | scale-guides/hobby-groups.md | templates/hobby-group-adr-template.md |
| Small team | Status, Date, Decision maker, Advisors | Add Confidence if uncertain. | scale-guides/small-teams.md | templates/small-team-adr-template.md |
| Business | Status, Date, Decision maker, Approvers, Stakeholders, Confidence, Cost impact | Full governance. Add Compliance notes. | scale-guides/business-projects.md | templates/business-adr-template.md |
If you don't know the scale, ask. Guessing produces either overengineered ceremony or missing accountability.
Load ONLY the guide and template for the scale you are working with. Do not load other scale files — each is self-contained.
Scope Sizing
Right-size the ADR to the scope of the decision.
| Scope | Use when | Expected size |
|---|---|---|
| Microscopic / local | A focused decision inside one component, module, or workflow | 3–6 short bullets, ~0.25 page |
| Mid-level / cross-cutting | A decision affects several modules or a repeated team workflow | ~1 page |
| Holistic / global / major | Changes system shape, platform direction, interfaces, or an expensive-to-reverse constraint | 1–2 pages with sharper tradeoff detail |
Scale by blast radius, reversibility, and downstream impact — not by importance theater.
Core Pattern
Each ADR answers these questions, in this order:
- What decision are we making?
- What context or forces made this decision necessary?
- Which serious alternatives were considered?
- Why did this option win?
- What consequences or tradeoffs follow from it?
- What would cause us to revisit or supersede it?
If the ADR cannot answer these questions, even briefly, the thinking is not done yet.
ADR Templates
Per-scale templates are in templates/. Each is purpose-built for its context — not just different metadata headers, but different structure, tone, and guidance:
templates/solo-adr-template.md— minimal, sections are optional, tone is self-directedtemplates/hobby-group-adr-template.md— lightweight, shared but informaltemplates/small-team-adr-template.md— professional, one decision-maker, async reviewtemplates/business-adr-template.md— full governance, required fields, quantified impact
Load only the template for the scale you determined above. Do not load templates for other scales.
Keep one ADR per file and one decision per ADR. Do not add sections just to look complete — include only fields that clarify the decision.
Writing Rules
- Put the decision near the top
- Write in plain, assertive language
- Capture rationale, not just outcome
- Name real alternatives, not strawmen
- State tradeoffs honestly
- Include concrete revisit triggers, not "review later"
- Keep accepted ADRs append-only; supersede with a new ADR instead of rewriting history
- Store ADRs near the code or docs they govern
Quick Reference
| Situation | Action |
|---|---|
| Decision will shape later implementation | Write an ADR |
| Reversing it would be annoying, risky, or expensive | Write an ADR |
| Multiple plausible options existed | Write an ADR |
| Tradeoff discussion matters more than the final label | Write an ADR |
| Just a task breakdown | Do not write an ADR |
| One tiny code tweak with no lasting design meaning | Do not write an ADR |
| Choice is still being explored | Keep designing |
| Document would just repeat the design doc | Link to the doc |
| Working solo — unsure if this matters | Ask the human partner |
Common Mistakes
| Mistake | Fix |
|---|---|
| Writing a design document instead of a decision record | ADRs capture the decision and rationale, not the full solution |
| Recording several decisions in one ADR | One decision per ADR. Split them. |
| Listing only the winner, hiding the alternatives | Name real alternatives and why they lost |
| Saying "we can revisit later" without triggers | Name concrete conditions |
| Solo project with committee-level ceremony | Use solo-scale metadata. Ask the human what they want. |
| Mixing implementation steps into the ADR | Action items belong in plans or tasks |
| Rewriting old ADRs instead of superseding them | Write a new ADR that supersedes the old one |
| Making microscopic decisions sound global | Right-size to actual blast radius |
| Auto-applying ADR template to solo project without asking | Ask first. Template is a tool, not a mandate. |
Rationalization Prevention
| Excuse | Reality |
|---|---|
| "This is too small to document" | If the tradeoff will matter later, a short ADR is enough |
| "We already have a design doc" | Design docs explain the solution; ADRs preserve the decision and rationale |
| "I will remember why we chose this" | Future-you usually remembers the winner, not the losing options |
| "I will just update the old ADR later" | Accepted ADRs should be superseded, not rewritten |
| "There were no real alternatives" | If nothing else was plausible, you probably do not need an ADR |
| "I should add rollout steps" | Action items belong in plans or tasks, not in the ADR |
| "I'm working alone — ADRs are overkill" | A 3-line ADR preserves rationale. Future-proofing, not ceremony. |
| "The template requires all metadata fields" | The template adapts to project scale. See Adaptive Metadata above. |
Red Flags — Stop and Fix
Metadata misuse:
- Auto-applying full governance metadata to a solo project without asking
- Using solo-level metadata for a regulated business decision
- Guessing the project scale instead of asking
Content problems:
- More than one real decision in the same ADR
- No rejected alternatives listed
- No tradeoffs or downsides acknowledged
- No concrete revisit trigger
- Status is unclear or missing
Scope problems:
- The document reads like a migration plan or implementation checklist
- Scope inflated far beyond the actual blast radius
Workflow
- Confirm the decision is settled. If still exploring, keep designing.
- Determine project scale. Solo? Hobby group? Small team? Business? If unsure, ask.
- For solo: ask the human partner whether to capture the decision and at what detail level.
- Right-size the record to local, cross-cutting, or global scope.
- Write one concise ADR for one decision, using metadata appropriate to the scale.
- Link supporting design material instead of copying it.
- If the decision changes later, write a new ADR that supersedes the old one.
Default Agent Behavior
When you and your human partner reach a meaningful decision, offer to capture it as an ADR — adapted to project scale.
- Non-solo: "We just made a meaningful design decision. Want me to capture it as an ADR so the rationale and tradeoffs do not get lost?"
- Solo: "We just made a design decision. Since you're working solo, would you like me to capture it as an ADR? I can keep it minimal or include full rationale — your call."
Offer an ADR when the decision is important enough to revisit later, likely to affect future choices, hard to reverse, or the result of real tradeoff discussion. Do not interrupt every tiny choice with ADR ceremony.
Supplemental Files
All supplemental files are in scale-guides/ (per-scale adaptation guidance) and templates/ (per-scale ADR templates). Load only the files for the scale you are working with — each is self-contained. The Adaptive Metadata table above lists the exact files per scale.