Reconstruction: Use when rough, mixed-language, or noisy input needs a short, repository-grounded English prompt. Own a routing-sufficient Intake contract; Skip routing-ready input or translation/editing.

Interleaving-first concurrency control for shared durable state. Use when designing, building, auditing, or diagnosing lost updates, double effects from concurrent requests, check-then-act races, duplicate submissions, optimistic vs pessimistic locking, version columns and ETags, unique-constraint arbitration, isolation anomalies like write skew, advisory and distributed locks, fencing tokens, leader election, or singleton work. Route job claim/lease design to durable-background-jobs, PostgreSQL lock contention as a bottleneck to postgres-performance, and constraint modeling to postgres-schema-design.

DigitalOcean benchmark: Use for authorized Go/PostgreSQL/HTTP measurements on ephemeral Droplets. Own paid-resource security, comparable evidence, artifacts, and cleanup; Skip local or undecided workloads.

boundary-first production integration with external HTTP providers. Use for design, build, diagnosis, or review of API clients, OAuth credentials, idempotency, deadlines/retries/rate limits, ambiguous outcomes, synchronization, webhooks, reconciliation, and provider migration. Own request identity through convergence and proof; hand public API, business/ledger, messaging, and PostgreSQL internals to their dedicated skills.

Chi transport: Use for router composition, middleware, OpenAPI wiring, fallbacks, CORS, labels, or routing review. Own chi decisions; Skip client API semantics, system topology, or general Go defects.

Data architecture: Use for authority, tenant identity, schema evolution, backfills, retention, projections, or datastore fit. Own truth/lifecycle; Skip runtime DB/cache, domain, recovery, or code placement.

Delivery platform: Use for CI/CD gates, release trust, drift, containers, migrations, rollout, or control review. Own delivery acceptance; Skip API, data, security, reliability, or pipeline implementation.

Distributed flow: Use for cross-service consistency, sagas, durable replay, ordering, compensation, redrive, or reconciliation. Own recovery; Skip local transactions, synchronization, lifecycle, or topology.

Domain invariants: Use for business terms, transitions, acceptance, violations, domain replay, or effect order. Own domain policy; Skip transport, data/cache mechanics, security, or test structure.

Go idiom: Use when changed Go risks errors, context, nil/zero, method sets, aliasing, or resource lifetimes. Own semantics; Skip readability, whole-diff structure, or package ownership.

Implementation ownership: Use for Go package/file/dependency, cleanup, proof placement, or boundary review. Own code placement; Skip system topology, domain policy, Go semantics, or local readability.

Go readability: Use when correct local Go is obscured by control flow, predicates, naming, or helpers. Own behavior-preserving simplification; Skip Go semantics, package ownership, or whole-diff overbuild.

Security: Use for trust boundaries, identity, authorization, tenant isolation, tokens, secrets, injection, or SSRF. Own policy; Skip non-security API, data, reliability, or placement.

Structural quality: Use for harsh whole-diff review or cross-file overbuild. Own abstraction cost, mixed responsibility, and missed deletion; Skip Go semantics, local readability, or architecture.

Pressure-test: Use when research/framing needs discriminating questions before spec/planning. Own hidden risks and next owner; Skip blank framing, ordinary review, or one approval question.

Evidence-first diagnosis of production backend incidents and regressions. Use when latency, errors, throughput, saturation, or correctness degrades in a running system and the cause is unknown: quantifying the symptom, mitigation-versus-diagnosis decisions, correlating deploys, config, flags, traffic, and data growth, localizing the failing hop with traces, metrics, and logs, falsifying causes, and proving recovery. After localization route the cause to postgres-performance, cache-engineering, reliable-messaging, durable-background-jobs, or external-api-integration; own the loop end to end when no dedicated skill applies.

Approval question: Use for one high-impact, hard-to-reverse, or challenged spec decision. Own read-only evidence/options/recommendation; Skip blank framing, whole-spec review, or low-impact clarification.

Problem frame: Use when a chosen feature/refactor needs a behavior delta, scope, constraints, assumptions, and readiness. Own that frame/next owner; Skip raw ideation, final design, or task breakdown.

Handoff integrity: Use for contested workflow-plan.md routing or an explicit challenge. Own a read-only anchored gap/next owner; Skip ordinary low-risk plans, authoring, or state changes.