| name | engineering-standard |
| description | Apply binding engineering constraints when implementing, fixing, debugging, designing, reviewing, refactoring, or migrating durable product code, APIs, handlers, workers, architecture, capabilities, DDD/Clean/Hexagonal boundaries, functional cores, contracts, storage, concurrency, security, tests, observability, or delivery-critical runtime paths. Use when code or runtime behavior is the artifact; use a specialist procedure when a whole observability, security, migration, or product contract is primary. |
Engineering Standard
Requirement: apply this standard when the task matches its scope.
Full progressive-disclosure body: references/full-standard.md.
Rule IDs: references/binding-predicates.md.
For code structure or architecture migration, also read:
When creating a project, adding a package, or upgrading a runtime, framework,
SDK, generator, plugin, or library, use dependency-version-selection. Query
the authoritative live release source; never select a version from model
memory, a stale template, or an old example.
When invariants span a broad generated state or input space, read
Generative verification.
When exact expected outputs are unavailable but relations or independent
implementations can falsify behavior, read
Oracle-free verification.
When a test, gate, alert, validator, or recovery control claims to catch a
failure, read
Control-effectiveness verification.
When diagnosing or fixing a defect, changing observable behavior, or
refactoring behavior whose current contract is not yet executable, read
Reproduction-driven engineering.
For critical concurrency, distributed coordination, authorization, or
irreversible-effect state spaces, read
Formal modelling and model checking.
When concurrent or offline replicated state needs an algebraic merge, read
CRDT convergence.
For a large shared capability catalog, concept scheme, faceted vocabulary, or
cross-domain taxonomy, use semantic-taxonomy-design; do not force identity,
hierarchy, and mutable labels into one tree path.
For a new or materially changed trust boundary, privileged action, data flow,
integration, or deployment exposure, use security-threat-modeling to produce
the security-design contract before or alongside implementation.
Method
- Load this package for durable engineering work on product code.
- Read references/full-standard.md for the complete standard.
- Before broad implementation, create or update the owning repository ADR for
a material durable decision about architecture, ownership, public contracts,
persistence, security/privacy, delivery semantics, or an enterprise default.
- Keep ordinary implementation detail governed by an accepted decision in
code, schemas, and tests; do not create ceremonial ADRs.
- Apply the strongest relevant subset and prefer executable evidence over
copying this body into product repositories.
- Procedure Skills should name this standard as a composition handoff when
durable product code or runtime behavior changes. Native runtimes select
matching Skills from metadata; the catalog has no dependency primitive, so
do not invent one or restate these predicates in every procedure.
Boundaries
- Does not grant deploy, credential, or MCP capabilities.
- Does not replace product-local technical ADRs.
- Does not auto-route as a procedure; it is binding policy text with separate conformance evidence.
When not to use
- Commercial packaging alone → commercial-decision class packages.
- Product-specific live coordination operations → the matching adapter Skill
and that product's current API or tool schema.
- One-off scripts with no durable product surface and no shared contract impact.
Composition and output
Apply this standard as constraints on the artifact requested by the user or
owned by the primary procedure. Do not emit a separate per-standard compliance
report merely because this Skill was injected.
Integrate only material obligations, deviations, evidence, and gaps into the
primary artifact or final status. When conformance assessment is itself the
requested job, produce one standalone domain conformance record from this
standard.
Guardrails
- Fail closed on secret material in source and logs.
- Never treat health/readiness probes as product capability proof.
- In development or without live compatibility/state risk, prefer a verified
one-step cutover. Use expand-contract only when a demonstrated live data,
compatibility, availability, or external-effect boundary requires it.
- Apply the canonical Capability-first architecture from the first durable
product commit; project size and migration cost are not exemptions.