Work-type skill
Cloud Infrastructure and Release
Govern cloud infrastructure, deployment, rollback, recovery, and release work.
Skill orientation
Use this panel to select and sequence the skill. The canonical source follows below.
- Purpose
- Govern cloud infrastructure, deployment, rollback, recovery, and release work.
- Category and sequence
- Work type; sequence 6
- Primary use
- Cloud, Deployment, Release, Reliability, Incident
- Required gates
- Classification, Foundation, Code Intelligence, Production, Proof
- Conditional gates
- smoke for deployed, provider-integration, provider-managed infrastructure, distribution-channel, or runtime behavior; domain pack for provider-managed infrastructure, distribution-channel, or specialized runtime
- Red Zone triggers
- Production Deploy, Testflight Upload, App Store Promotion, Dns, Traffic, Secrets, Cloud Resource, Failover, Chaos, Charge, Refund, Void, Financial Reconciliation, Payment Provider Configuration, Customer Communication
- Next route
- Return to the active lifecycle route
- Source identifier
skills/valdris-platform-release/SKILL.md
Valdris Platform And Release
- Load the adapter's environment, provider, branch, deployment, and approval rules; validate the intake, classification, route, and code-intelligence artifacts.
- Resolve the route-required Layer 0 foundation assessment before changing infrastructure or release behavior.
- Produce a service and dependency map with owners, regions, data boundaries, and failure modes.
- Require reviewable IaC or configuration diffs; reject dashboard-only claims without exported evidence.
- Define workload-specific SLOs, error budgets, capacity assumptions, cost ceilings, rollback triggers, RTO, and RPO.
- Validate CI provenance, staged promotion, health checks, telemetry, alerts, rollback, and restore. Require smoke only when the route identifies deployed, provider-integration, platform-native, or runtime behavior. A prototype-to-production move requires
uash.promotion-receipt.v1with a distinct production build and semantic proof; prototype evidence cannot be relabeled as production evidence. - Exercise failure or recovery only inside the approved blast radius.
- Classify all affected production assurance domains and hard dependencies; treat conditional dependencies as applicability questions, not automatic expansion.
- For authoritative claims, execute proof in the commissioned immutable executor and advance the rollback-resistant bridge head with a provider-backed compare-and-swap receipt. The agent may not inherit ambient secrets, weaken network/resource policy, self-enroll a trust key, or accept a stale head.
- If the current host cannot run platform-native proof, keep that stopping condition open and route it to an approved runner; never substitute source inspection for a signed build, device test, provider smoke, restore, or failover result.
Stop for human approval before production deploys, TestFlight/App Store uploads or promotions, traffic/DNS changes, secrets, IAM, data restore, failover, chaos tests, spend commitments, payment-provider changes, charges/refunds/voids/reconciliation, or customer communications.
Availability without correctness is not recovery; degraded AI fallbacks must also pass quality and safety gates.
