/blueprint:adr-validate

Validate Architecture Decision Records for relationship consistency, reference integrity, and domain conflicts.

Usage: /blueprint:adr-validate [--report-only]

When to Use This Skill

Context

• ADR directory exists: !find . -path '*/docs/adrs' -maxdepth 2 -type d
• ADR count: !find . -path '*/docs/adrs/*' -name "*.md" -type f
• Domain-tagged ADRs: !find . -path '*/docs/adrs/*' -maxdepth 3 -name '*.md' -exec grep -l "^domain:" {} +
• Flag: !echo "${1:---}"

Parameters

Parse $ARGUMENTS:

• --report-only: Output validation report without prompting for fixes
  • Default: Interactive mode with remediation options

Execution

Execute complete ADR validation and remediation workflow:

Step 1: Discover all ADRs

1. Check for ADR directory at docs/adrs/
2. If missing → Error: "No ADRs found in docs/adrs/"
3. Parse all ADR files: ls docs/adrs/*.md
4. Extract frontmatter for each ADR: number, id, created, modified, status, domain, supersedes, superseded-by, extends, relates-to

Step 2: Validate reference integrity

For each ADR, validate:

1. supersedes references: Verify target exists, target status = "Superseded", target has reciprocal superseded-by
2. extends references: Verify target exists, warn if target is "Superseded"
3. related references: Verify all targets exist, warn if one-way links
4. self-references: Flag if ADR references itself
5. circular chains: Detect cycles in supersession graph
6. Cross-workspace references (v3.3.0+, manifests with workspaces.role): Recognise these reference forms in supersedes/extends/related fields:
   • ADR-NNN — local to the current workspace (existing behaviour).
   • <workspace-path>/ADR-NNN — points into a sibling/child workspace. Resolve by reading <workspace-path>/docs/adrs/ from the monorepo root. Warn if the workspace is not listed in root workspaces.children.
   • /ADR-NNN — points at the monorepo root's ADR set. Resolve using the manifest's workspaces.root_relative_path (for child manifests) or the current directory (for root manifests). Unresolved cross-workspace refs are reported as warnings (not errors) so they do not block validation during migration.

See REFERENCE.md for detailed checks.

Step 2b: Detect ADR-number collisions and index drift

ADR numbers are chosen at branch time but only claimed at merge time, so two in-flight ADR PRs can pick the same number and both land (issue #1585 — the FVH infrastructure #2015 collision where two ADRs both claimed 0038). Run the deterministic guard:

bash ${CLAUDE_SKILL_DIR}/scripts/check-adr-numbers.sh --project-dir "$(pwd)"

It emits the structured STATUS= / ISSUE_COUNT= convention and reports four classes (see REFERENCE.md):

• duplicate_adr_number (ERROR) — two files claim the same number, by NNNN-title.md filename or by a frontmatter id: ADR-NNNN claim. The message names the source (basename / frontmatter), so a claim from a file with no number in its name is legible.
• adr_number_collision (ERROR) — a working-tree ADR claims a number a different file already holds on the base ref (origin/main) — the pre-merge parallel-PR case, caught before the second PR merges.
• adr_registry_mismatch (ERROR) — a file claims ADR-NNNN but the manifest id_registry maps that number to a different path. The registry is the only arbiter resolving id → path unambiguously: two files can both say ADR-0016, only one can be registered. Silent with no manifest / no id_registry.
• adr_missing_index_row (WARN) — an ADR is missing from the directory's README index (how the 0038 collision went unnoticed for a week). Repair via Step 2c instead of re-warning.

Multiple ADR directories (issue #2129): the guard scans a directory set — the real ADR-0016 collision had its claimants in docs/adrs/ and docs/blueprint/adrs/, which a single-directory scan cannot see. Precedence: repeatable --adr-dir <dir>, else validation.adr_dirs from the shared validation manifest block (issue #2128, read via blueprint-plugin/scripts/get-validation-config.sh — do not add a second config mechanism), else the default docs/adrs then docs/adr order. It emits ADR_DIRS= (TAB-separated) and ADR_DIR_COUNT=; ADR_DIR=, ADR_COUNT= and INDEX= keep their #1585 meanings.

Fold findings into the Step 4 report. STATUS=ERROR is a blocking collision: renumber the newer ADR to the next free number (updating both its filename and its frontmatter id:), then regenerate the index (Step 2c).

Step 2c: Regenerate the ADR index

adr_missing_index_row is repairable — the hand-maintained table is what rotted. Rebuild it from frontmatter:

# CI / read-only gate: non-zero exit when the on-disk index is out of date
bash ${CLAUDE_SKILL_DIR}/scripts/generate-adr-index.sh --project-dir "$(pwd)" --check

# Repair: rewrite the table in every resolved ADR directory
bash ${CLAUDE_SKILL_DIR}/scripts/generate-adr-index.sh --project-dir "$(pwd)"

Only the block between two marker comments is rewritten, so prose around it survives:

<!-- ADR-INDEX:START -->
| ADR | Title | Status |
| --- | ----- | ------ |
| [ADR-0016](0016-in-engine-ui.md) | In-engine UI | Accepted |
<!-- ADR-INDEX:END -->

One index per directory (--index-file overrides the README.md default), over the same directory set and widened number collection as Step 2b — so an ADR numbered only in frontmatter still gets a row. --check states per directory:

Write mode reports written / unchanged / markers_inserted / index_created. When markers land in a README that already had a hand-written table, delete that table — the generated block is authoritative.

Step 3: Analyze domains

1. Group ADRs by domain field
2. For each domain with multiple "Accepted" ADRs → potential conflict flag
3. List untagged ADRs (not errors, but recommendations)

Step 4: Generate validation report

Compile comprehensive report showing:

• Summary: Total ADRs, domain-tagged %, relationship counts, status breakdown
• Reference integrity: Supersedes, extends, related status (✅/⚠️/❌)
• Errors found: Broken references, self-references, cycles
• Warnings: Outdated extensions, one-way links
• Domain analysis: Conflicts and untagged ADRs

Step 5: Handle --report-only flag

If --report-only flag present:

1. Output validation report from Step 4
2. Exit without prompting for fixes

Step 6: Prompt for remediation (if interactive mode)

Ask user action via AskUserQuestion:

• Fix all automatically (update status, add reciprocal links)
• Review each issue individually
• Export report to docs/adrs/validation-report.md
• Skip for now

Execute based on selection (see REFERENCE.md).

Step 7: Update task registry

Update the task registry entry in docs/blueprint/manifest.json:

jq --arg now "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
  --arg result "${VALIDATION_RESULT:-success}" \
  --argjson processed "${ADRS_VALIDATED:-0}" \
  '.task_registry["adr-validate"].last_completed_at = $now |
   .task_registry["adr-validate"].last_result = $result |
   .task_registry["adr-validate"].stats.runs_total = ((.task_registry["adr-validate"].stats.runs_total // 0) + 1) |
   .task_registry["adr-validate"].stats.items_processed = $processed' \
  docs/blueprint/manifest.json > tmp.json && mv tmp.json docs/blueprint/manifest.json

Where VALIDATION_RESULT is "success", "{N} warnings", or "failed: {reason}".

Step 8: Report changes and summary

Report all changes made:

• Updated ADRs (status changes, added links)
• Remaining issues count
• Next steps recommendation

Agentic Optimizations

For validation rules, remediation procedures, and report format details, see REFERENCE.md.