Agent-design-language adr-curator
Draft source-grounded Architecture Decision Record candidates from CodeBuddy review packets, architecture reviews, findings, migration notes, docs, PR notes, or repo evidence with status, context, decision, consequences, supersession links, validation notes, and explicit proposed/accepted boundaries while stopping before accepting decisions or mutating repositories.
git clone https://github.com/danielbaustin/agent-design-language
T=$(mktemp -d) && git clone --depth=1 https://github.com/danielbaustin/agent-design-language "$T" && mkdir -p ~/.claude/skills && cp -r "$T/adl/tools/skills/adr-curator" ~/.claude/skills/danielbaustin-agent-design-language-adr-curator && rm -rf "$T"
adl/tools/skills/adr-curator/SKILL.mdADR Curator
Draft reviewable Architecture Decision Record candidates from CodeBuddy packets, architecture reviews, findings, migration notes, docs, PR notes, and repo evidence. This skill is a curation lane, not a decision authority.
It may identify candidate decisions, draft proposed ADR text, preserve evidence, and connect supersession links. It must not accept an ADR, change architecture, create tracker items, open PRs, or mutate customer repositories unless a separate approved implementation workflow explicitly authorizes that work.
Quick Start
- Confirm the bounded source:
- CodeBuddy review packet
- architecture review artifact
- synthesis artifact
- findings file
- migration notes or docs path
- repo path with existing ADR evidence
- Prefer packet artifacts when available:
evidence_index.jsonrepo_inventory.jsonrun_manifest.json- specialist architecture or dependency review artifacts
- Run the deterministic curation scaffold when local access is available:
scripts/curate_adrs.py <source-root> --out <artifact-root>
- Review the emitted ADR candidates and tighten the status, context, decision, consequences, alternatives, and supersession links.
- Stop before accepting, publishing, committing, or opening tracker/PR work.
Focus
Prioritize:
- candidate decisions already implied by code, docs, migrations, reviews, or PRs
- decisions that need durable explanation before follow-up implementation
- status clarity: proposed, accepted-existing, superseded-existing, or deferred
- context, decision, consequences, alternatives, validation notes, and evidence
- supersession and replacement relationships between decisions
- clear distinction between observed repo truth and proposed new ADR language
Defer primary ownership of these areas:
- finding original architecture defects:
repo-architecture-review - planning executable architecture checks:
architecture-fitness-function-author - creating follow-up issues:
finding-to-issue-planner - final report synthesis:
or report writer skillsrepo-review-synthesis - editing ADR files in a repository: implementation issue workflow
Required Inputs
At minimum, gather:
repo_root- one concrete target:
target.review_packet_pathtarget.architecture_review_artifacttarget.findings_filetarget.migration_notes_pathtarget.target_path
Useful additional inputs:
artifact_rootadr_status_policysupersession_policyvalidation_modeexisting_adr_dir
If there is no bounded decision source or evidence, stop and report
blocked.
Workflow
1. Establish Scope
Record:
- source artifacts reviewed
- repo/path scope
- existing ADR directory, if known
- whether accepted ADRs already exist
- whether proposed ADR drafts are allowed
Do not widen one candidate decision into a whole architecture history project.
2. Extract Candidate Decisions
Look for:
- explicit ADR files or headings
,Status
,Context
,Decision
, andConsequences
fieldsSupersedes- architecture-review candidate ADR sections
- migration notes that imply durable decisions
- repeated review findings that need a policy decision before automation
- PR notes that changed boundaries, providers, persistence, runtime, or lifecycle behavior
For each candidate, decide whether it is:
: a draft decision needs human acceptanceproposed
: an existing source already records acceptanceaccepted_existing
: an existing source records supersessionsuperseded_existing
: evidence is too weak or the decision boundary is uncleardeferred
Default to
proposed when status is not explicit.
3. Draft ADR Candidates
Each ADR candidate should include:
- ADR id
- title
- status
- source evidence
- context
- decision
- consequences
- alternatives considered
- supersession links
- validation notes
- approval boundary
Keep proposed ADRs visibly proposed. Do not convert a proposed draft into an accepted decision without explicit source evidence.
4. Emit Handoffs
Recommended handoffs:
for executable checks after a decision is approved or stable enough to planarchitecture-fitness-function-author
for follow-up issue candidatesfinding-to-issue-planner- implementation issue workflow for committing accepted ADR files
when evidence is too weak to draft safelyrepo-architecture-review
Do not invoke downstream skills automatically unless the operator explicitly asks for that follow-on execution.
Output Expectations
Default output should include:
- ADR candidate catalog
- proposed ADR drafts
- accepted or superseded existing decisions, if any
- deferred decision candidates
- supersession map
- validation notes
- approval boundary
- residual decision risk
Use
references/output-contract.md and the shared suite contract in
adl/tools/skills/docs/MULTI_AGENT_REPO_REVIEW_SKILL_SUITE.md.
Stop Boundary
Stop after producing the ADR candidate packet.
Do not:
- accept, reject, supersede, or publish architecture decisions
- edit ADR files, docs, tests, CI, code, policy files, issues, or PRs
- mutate customer repositories
- create issues or PRs
- claim a decision has been accepted when only a proposed draft exists
- use network or paid services
CodeBuddy Integration Notes
This skill consumes CodeBuddy packet artifacts, review outputs, and repo evidence. It produces ADR candidate packets that can feed architecture review, fitness-function planning, issue planning, implementation work, or final customer reports after human review.
Deferred automation:
- language-specific architecture decision inference
- automatic ADR file placement and numbering
- repo-specific ADR template adoption
- cross-run ADR history and supersession graph maintenance