Audit Contract Changes Across All System Boundaries

Cross-Platform Contract Propagation Audit checks whether a shared field, enum, status, capability, or feature flag keeps the same meaning across its entire propagation path. It starts from the business invariant and follows persistence, migrations, domain mapping, service computation, API and event projections, client models, presentation, analytics, and tests. The workflow is read-only: it identifies gaps and the smallest verification or repair set needed to change the release verdict, rather than implementing changes.

The audit first defines observable states such as a missing property, explicit null, false or zero, true or non-zero, and an unknown enum value. For each state it records expected behavior, compatibility requirements, ownership, and rollout conditions. Optional, nullable, and default-false declarations are not assumed to mean the same thing. A type declaration establishes shape, while runtime mapping and behavior need their own evidence.

Use this skill when a contract change spans a database, backend, Web, Android, iOS, jobs, events, or analytics, especially when existing records, older clients, or a default-off rollout must continue working. It is useful when one endpoint looks complete but list views, alternate entry points, generated models, background consumers, offline caches, or administrative surfaces may still omit the change.

The audit does not establish that the business rule itself is correct or discover every security, performance, or product-design defect. Static repository evidence cannot prove live configuration, deployed schema state, generated-code freshness, or unexercised client behavior. Unavailable private services, remote flags, analytics schemas, and older released clients remain unknown. Implementation and runtime testing require separate authorization; production changes must not be made merely to fill an evidence gap.

Inputs include the semantic contract, relevant repository revision, scoped components, schemas, mappings, flags, analytics definitions, and inspectable test evidence. The propagation graph includes every material producer, transformation, and consumer. Trace this propagation path:

source of truth
  -> persistence and migration
  -> domain model and mapper
  -> service or policy computation
  -> every API, event, cache, and job projection
  -> generated or handwritten client model
  -> client state and presentation logic
  -> analytics and operational observability
  -> tests, rollout, and rollback checks

For each edge, findings cite files, symbols, schema names, and tests and assign proven, partial, missing, conflict, unknown, or not_applicable. Proven requires producer-consumer agreement with direct evidence and relevant coverage. A not-applicable layer needs a reason; conventions or framework defaults are not proof. Negative searches include their scope and revision so absence claims can be reproduced.

The high-risk review covers existing-data defaults, backfill, nullability, rollback, mixed-version reads and writes, coercion, enum fallbacks, validation, and serialization symmetry. It also checks list/detail DTOs, events, caches, jobs, search indexes, SDKs, API versions, client decoding, cached payloads, and release builds. Rollout evidence includes flag defaults, evaluation points, cohorts, a kill switch, and disagreement between stored values and flags. Analytics must preserve enough contract and version context to join offered, rendered, attempted, succeeded, and failed actions.

A state-by-path matrix records expected results, evidence, and status for defaults, enabled and disabled values, flags on and off, alternate endpoints, and current and older clients. Unexecuted cells stay unknown. Explicit required gates that are missing, conflicting, or unknown block release, as does inadequate rollback containment. Inconclusive applies only when the release contract itself is absent or ambiguous. Findings separate proven gaps from implementation suggestions and redact production records, credentials, user identifiers, and sensitive payloads.