Find Contract Drift

/find-contract-drift

Run an advisory contract audit for configured product-workflow targets. This skill reuses /find-frontend-contract-drift for boot-global contracts, then adds DOM, endpoint, and template-to-JS checks that are specific to the scanned workflow files.

Default scope is not hard-coded to /sites. With no positional paths, scripts/run.py passes None into detect(), and product_health.expand_paths() scans workflow_targets(project_root) from .engineering/docs/product-workflows.md (## Targets). If that descriptor or section is absent, the default workflow target set is empty. To scan a concrete surface, pass explicit positional paths after scripts/run.py. On no-path runs, the detector may also add boot-contract coverage from workflow_template_roots(project_root) and templates/core/includes; pass explicit paths when you need exact scan scope.

How success is judged

• The runner exits 0 and prints the scan directory it wrote.
• reports/find-contract-drift/scan-<UTC>/detections.jsonl, report.md, findings.json, and latest exist under the selected --project-root.
• The target scope is explicit in the command or comes from workflow_targets(project_root); do not describe a no-descriptor run as a /sites scan.
• Findings are grouped by contract detector bands and are treated as advisory triage input; high-confidence DOM/export/endpoint findings may feed /fix-workflow or extract-workflow-registry.

Pipeline

.venv/bin/python .claude/skills/find-contract-drift/scripts/run.py <paths...>

<paths...> are positional path or glob arguments. Omit them only when the host repo has declared ## Targets in .engineering/docs/product-workflows.md.

The runner writes:

• reports/find-contract-drift/scan-<UTC>/detections.jsonl
• reports/find-contract-drift/scan-<UTC>/report.md
• reports/find-contract-drift/scan-<UTC>/findings.json
• reports/find-contract-drift/latest

Detector Bands

• implicit_template_global and related boot-contract findings imported from /find-frontend-contract-drift.
• missing_endpoint_key: JS calls SiteConfigCore.siteEndpoint*("key") for a key absent from SiteWorkflowRegistry.SITE_ENDPOINTS or SITE_ENDPOINT_TEMPLATES.
• missing_js_export: a template calls window.Namespace.method() but the scanned JS namespace export does not expose that method.
• missing_dom_id: JS reads an ID that is absent from scanned templates.
• missing_data_selector: JS queries a data-* selector absent from scanned templates.

Treat results as triage input, not a chore pile. High-confidence new findings are candidates for cleanup; noisy or intentional findings should be captured as fixture refinements or accepted tradeoffs.

When things go sideways