Architecture Decision Record (ADR) management skill. Creates, maintains, and enforces architectural decisions with anti-rationalization guards and named verification gates. Drop into any project to give an AI coding agent a shared, enforceable ADR workflow.

2026-07-073

init

rvdbreemen/adr-kit

One-shot project bootstrap for adr-kit. Hooks the kit into CLAUDE.md (via a slim stub + a copy of templates/adr-kit-guide.md → .claude/adr-kit-guide.md), runs bin/adr-audit to enumerate decision-shaped artefacts in source + documentation, walks the user through batch approval to generate Accepted ADRs via the adr-generator subagent, installs the pre-commit hook, and finally lints. Idempotent across re-runs. User-invocable only — this is a side-effecting operation.

2026-07-053

setup

rvdbreemen/adr-kit

One-time project setup for adr-kit. Hooks `CLAUDE.md` (slim stub with @-import) and drops the canonical guide at `.claude/adr-kit-guide.md`. v0.11-style inline `## ADR Kit Rules` sections are detected and left untouched (run `/adr-kit:upgrade` to migrate them). Idempotent across re-runs. The lighter cousin of `/adr-kit:init` — `setup` does not run a codebase audit or install the pre-commit hook.

2026-07-053

install-hooks

rvdbreemen/adr-kit

Install or uninstall the adr-kit pre-commit hook in the current project. Copies templates/githooks/pre-commit into .githooks/pre-commit, makes it executable, and runs `git config core.hooksPath .githooks`. Idempotent. Used internally by /adr-kit:init and /adr-kit:upgrade; also exposed standalone for users who want to add or remove the hook independently.

2026-06-123

guardian

rvdbreemen/adr-kit

ADR-set health sweep for adr-kit. Runs the due health tier(s) — cheap (drift + stale + lint) and/or LLM (suggest + audit) — applies mix-by-finding-type responses, and stamps the state file when done. Invoke after seeing an [adr-guardian] ... DUE block at session start, or on demand to run a full health sweep. Accepts optional arg: cheap | llm | all. LLM tier always asks before spending.

2026-06-123

rvdbreemen/adr-kit

Shows the dependency graph for one Architecture Decision Record. Give it an ADR id (e.g. "ADR-007") and it lists outbound edges (Related Decisions entries, Supersedes claims, Superseded by / Amended by status refs) and inbound edges (every other ADR that references it, with the reference kind), flagging dangling references to ADRs that do not exist. Read-only and safe to call from parallel subagents. Invoke before superseding or retiring an ADR, or whenever you need to know what depends on a decision.

2026-06-123

name	migrate
description	Guided rewrite of legacy-shaped ADRs into the canonical-seven-section template enforced by /adr-kit:lint. Promotes inline status / date lines to a
argument-hint	[file or directory; defaults to docs/adr/]
disable-model-invocation	true
allowed-tools	["Read","Edit","Glob","Grep"]

adr-kit migrate

You are running /adr-kit:migrate. The user wants to rewrite one or more legacy-shaped ADRs into the canonical template that /adr-kit:lint enforces. This is a write skill: you propose targeted structural edits, the user confirms, then you apply them.

Inputs

No argument: migrate ADRs in docs/adr/ (the whole tree).
A directory path: migrate ADRs under that directory.
A file path: migrate that one file.

If the path does not exist or contains no ADR files, say so plainly and stop.

Cardinal rules

Read-then-confirm. Always read every target file before proposing edits. Always show the proposed restructure (file by file, summarised) and ask for explicit confirmation before calling Edit. Silent writes are forbidden.
No fabrication. If a canonical section is missing AND the source contains no inline content that maps to it, do NOT invent content. Leave a  placeholder so a human author can fill it in later.
Preserve content. The migration restructures shape, not substance. Bullet points, prose, examples, and links from the source must appear in the target unchanged. Reorder; do not rewrite.
Idempotent. Running migrate on an already-canonical ADR is a no-op (the skill detects "no missing sections" and skips the file).
Skip files that opt out. A file with  is left untouched. A file with  gets a warning ("this file is currently in advisory mode; migrating will make the marker meaningless") and a confirmation prompt.

Workflow

Step 1: read the policy

Look for docs/adr/.adr-kit.json (relative to project root, or the directory passed). If template.required_sections is set, target that list. Otherwise target the canonical seven:

## Status
## Context
## Decision
## Alternatives Considered
## Consequences
## Related Decisions
## References

Order matters. The migration must end with sections in the configured order.

Step 2: classify each input file

For each file, determine:

Already canonical (no missing sections, sections in the right order) -> skip, report as "no changes needed".
Marker says skip -> skip, report as "skipped per marker".
Marker says advisory -> ask "advisory marker is on this file; migrate anyway?"
Migratable -> identify which patterns apply (see next section).

Step 3: identify patterns

The skill body below documents the patterns observed in real-world legacy ADRs. Apply each pattern that fits, in this order. Do not invent new transformations.

Pattern A: inline status promotion

Source:

# ADR-NNN Title

**Status:** Accepted
**Date:** 2026-04-25
**Supersedes:** ADR-XXX (optional)

Target:

# ADR-NNN Title

## Status

Accepted, 2026-04-25. Supersedes ADR-XXX (optional).

If only **Status:** exists with no **Date:**, the Status section reads Accepted (without date).

Pattern B: alternatives inside Context

Source has a ### Alternatives considered: (or ### Alternatives considered and rejected) heading nested inside ## Context. Target: lift the entire block out, change ### to ## , and place the new top-level ## Alternatives Considered heading between ## Decision and ## Consequences. Preserve content verbatim; only the heading level and position change.

Pattern C: alternatives inside Consequences

Same as Pattern B but the source nests alternatives inside ## Consequences. Target: same restructure, place between Decision and Consequences.

Pattern D: Related to Related Decisions + References split

Source: a ## Related section that mixes ADR/TASK references with file paths, PR links, vendor doc URLs, internal docs.

Target: rename to ## Related Decisions. Move pure-external references (files, URLs, PRs that are not ADR or TASK identifiers) to a new ## References section that follows. Keep ADR-NNN and TASK-NNN entries in ## Related Decisions.

Heuristic for splitting:

Lines starting with ADR-, TASK-, or referencing those identifiers -> ## Related Decisions.
Lines starting with backticks (file path), URL, or PR /Issue references -> ## References.
When ambiguous, leave in ## Related Decisions (safer default).

Pattern E: missing References with no inline content

If after Pattern D there are no external references to populate ## References, create the section with a placeholder:

## References

<!-- TODO: populate from inline citations or external sources cited in the body. -->

Never invent references. The placeholder makes it clear to a human reviewer that this is a known gap.

Pattern F: missing Alternatives Considered with no source

If the source legitimately has no alternatives discussion anywhere, create the section with a placeholder:

## Alternatives Considered

<!-- TODO: document at least 2 alternatives that were considered and rejected, with reasoning. -->

This is a real gap a human should fill, but the skill must not fabricate.

Pattern G: MADR mapping

Source is a MADR-shaped ADR (madr.github.io): YAML frontmatter carrying status: / date:, plus ## Context and Problem Statement, ## Considered Options, ## Decision Outcome, and often ## Pros and Cons of the Options and ### Positive Consequences / ### Negative Consequences subsections. bin/adr-audit flags these files as template_profile: madr.

Mapping (content moves verbatim; only headings and position change):

MADR source	Canonical target
frontmatter `status:` + `date:` (or inline `* Status:` lines)	`## Status` section, e.g. `Accepted, 2026-04-02.` Drop the frontmatter block afterwards.
`## Context and Problem Statement` (+ `## Decision Drivers` bullets, folded in below the problem statement)	`## Context`
`## Considered Options` + `## Pros and Cons of the Options`	`## Alternatives Considered`: one bullet per option with its pro/con summary as the rejection or selection reasoning. The chosen option stays listed, marked as chosen.
`## Decision Outcome` ("Chosen option: ..., because ...")	`## Decision` (the chosen option and its justification)
`### Positive Consequences` / `### Negative Consequences` (or "Good, because" / "Bad, because" bullets under Decision Outcome)	`## Consequences` with `Positive:` / `Negative:` lists
`## Links` / `## More Information`	split per Pattern D: ADR identifiers to `## Related Decisions`, external links to `## References`

If the source has no Related Decisions or References content, apply Patterns E / F TODO placeholders. Never invent rejection reasons that the Pros and Cons section does not state; if an option has no documented cons, carry it over with .

Pattern H: Nygard lift

Source is a Nygard / adr-tools ADR: exactly ## Status, ## Context, ## Decision, ## Consequences, nothing else. bin/adr-audit flags these files as template_profile: nygard.

The four sections map 1:1 onto their canonical namesakes; their content is preserved verbatim. The three sections Nygard does not have are created with TODO placeholders, in canonical order:

## Alternatives Considered between Decision and Consequences, via Pattern F's placeholder.
## Related Decisions after Consequences, with - None. unless the Status or Context text references other ADRs (e.g. "Supersedes ADR-9"), in which case those references move here.
## References last, via Pattern E's placeholder.

Nygard Status sections often contain supersession links ("Superseded by [ADR-12]"). Keep them in ## Status as prose and mirror the ADR identifier into ## Related Decisions.

Step 4: present the plan

Before applying any edit, show the user a per-file summary:

Proposed migration plan (3 files):

ADR-007-timer-based-task-scheduling.md
  Pattern A: inline `**Status:**` -> `## Status` heading
  Pattern D: `## Related` -> `## Related Decisions` + new `## References`

ADR-029-simple-xhr-ota-flash.md
  Pattern A: inline `**Status:**` -> `## Status` heading
  Pattern F: missing `## Alternatives Considered`, will create with TODO placeholder

ADR-058-nonblocking-pic-command-response.md
  Pattern A: inline `**Status:**` -> `## Status` heading
  Pattern F: missing `## Alternatives Considered`, will create with TODO placeholder
  Pattern E: missing `## References`, will create with TODO placeholder
  No `## Related Decisions` content found in source; will create with `- None.`

Confirm to apply (y/n)?

If the user declines, stop without writing.

Step 5: apply edits

After confirmation, apply each transformation via Edit. One Edit per logical change, so the diff is reviewable. Report what was changed per file.

Step 6: post-migration verification

After all edits, suggest the user run /adr-kit:lint <path> to confirm the migrated files now PASS strictly. Do NOT run lint yourself: that is a separate skill the user invokes.

What you do not do

You do not modify the body content of sections. Headings move; prose stays.
You do not auto-fabricate Alternatives Considered or References content. Use TODO placeholders.
You do not rename ADR files (Consistency-gate filename FAILs). Out of scope.
You do not edit ADRs that already PASS strict.
You do not invoke /adr-kit:lint after migration. The user decides when to verify.

Reporting format

Single-file migration:

ADR-007-timer-based-task-scheduling.md migrated.
  Applied: Pattern A (Status promotion), Pattern D (Related split).
  Run /adr-kit:lint on this file to verify.

Directory migration:

Migrated 3 of 80 candidate files. 1 already canonical (skipped). 76 deferred (no patterns matched, manual review needed).

Applied:
  ADR-007 (A, D)
  ADR-029 (A, F)
  ADR-058 (A, F, E, no Related content)

Skipped:
  ADR-022 (already canonical)

Deferred (manual review): 76 files.
  Reason: complex shape that did not match any of patterns A through H. Inspect by hand.

Run /adr-kit:lint docs/adr/ to confirm overall result.

The aggregate's bottom line tells the user one concrete next step (run lint), never invents a status the migration did not actually achieve.

Edge cases

Multiple inline metadata lines: source has **Status:**, **Date:**, **Supersedes:**, **Amended by:**, etc. Fold all into the new ## Status section as a comma-separated sentence. Order: Status, date, supersedes/amended.
Empty Related section: source has ## Related with no body or only whitespace. Target: ## Related Decisions with - None. body.
Anchor comments inline at top: source has a Renumbered from ADR-XXX ... line before **Date:**. Fold into the new ## Status section as a trailing sentence: "Renumbered from ADR-XXX on YYYY-MM-DD to resolve duplicate numbering. Content unchanged."
Body has Markdown that confuses heading-detection: e.g. a ## inside a fenced code block. The skill treats only headings outside code fences as canonical sections. If unsure, ask the user.
Source uses ## Pros and Cons or ## Decision drivers: if the file matches the MADR shape, Pattern G consumes these (## Pros and Cons of the Options feeds ## Alternatives Considered, ## Decision Drivers folds into ## Context). Otherwise do not rename these; they are project-specific and the user can address custom subsections in a follow-up pass.

Anti-patterns to refuse

If the migration would require any of these, refuse and explain:

Renaming ADR files. Out of scope; surfaces commit / cross-ref breakage.
Modifying body prose ("the Decision section reads better if reorganised..."). Restructure shape only.
Fabricating Alternatives or References. TODO placeholder is the answer.
Skipping the user confirmation. Read-then-confirm is non-negotiable.
Running /adr-kit:lint automatically after migration. The user decides.