Staged Delivery Observability
Canonical OpenSpec capability for Staged Delivery Observability.
Document layout: Feature capability layoutconforms
## Purpose## Requirements
staged-delivery-observability Specification
Purpose
Require CI gates before promotion, digest-addressed artifact promotion, distinct staging and production environments, and end-to-end delivery traceability.
Requirements
Requirement: CI blocks unsafe integration
Pull-request and main-branch workflows SHALL require relevant format, type, unit, integration, conformance, OpenSpec, Compose, and security gates before artifacts may be promoted.
Scenario: A required smoke or conformance gate fails
- GIVEN an artifact was built successfully
- WHEN any required validation gate fails
- THEN no staging or production deployment job can start
Requirement: Artifacts are built once and promoted by digest
The delivery pipeline SHALL publish immutable SHA-addressed artifacts with SBOM, provenance, vulnerability results, and signatures, then SHALL deploy the exact same digest manifest to staging and production without rebuilding or substituting mutable tags.
Scenario: A release is promoted
- GIVEN staging passed for a signed manifest
- WHEN production approval is granted
- THEN production receives the identical image digests verified in staging
Requirement: Staging and production are distinct protected environments
The deployment system SHALL implement automatic staging deployment from main and protected production promotion with environment-specific secrets, URLs, policy gates, health checks, and rollback targets.
Scenario: Production health check fails
- GIVEN a previous healthy manifest exists
- WHEN post-deployment smoke or SLO checks fail
- THEN deployment is marked failed and rollback restores the previous manifest
Requirement: Delivery is traceable end to end
CI and deployment jobs SHALL emit a shared correlation identifier and OpenTelemetry-compatible trace context across workflow, build, registry, deployment API, service startup, smoke checks, and rollback reporting.
Scenario: Operator investigates a deployment
- GIVEN a GitHub Actions run identifier
- WHEN the operator follows its deployment evidence
- THEN the exact commit, OpenSpec revision, artifact digests, environment, deployment status, smoke results, and trace identifier are available
Requirement: Production failures are never suppressed
Production secret audits, deployment API calls, status polling, smoke checks, and release gates MUST fail closed; missing credentials or non-success responses MUST NOT be converted to warnings or ignored exit codes.
Scenario: OpenBao credentials are unavailable
- GIVEN a production deployment was requested
- WHEN required secret-audit credentials are missing
- THEN the workflow stops before deployment with a failing status