magnus919/agent-skills

production-readiness

- Define risk-scaled evidence packets and go/no-go/defer/exception launch gates with accountable owners.

Ver código-fonte
Documento original do Skill

Renderizado do repositório de origem, preservando títulos, exemplos, código, tabelas, links e imagens.

Production Readiness

Assemble cross-domain evidence into a launch decision. This skill defines the minimum production evidence packet by risk class, maps every evidence category to a named source or an explicit missing-evidence outcome, and produces one of four launch decisions with an accountable owner. It does not duplicate release checklists, does not produce a universal risk score, and routes detailed checks to the existing specialist skills.

When to load this

Load when any of these triggers matches:

  • A service, feature, or change is approaching a launch decision and needs a

production-readiness review.

  • Evidence from multiple domains (ownership, SLOs, security, data, support, cost,

observability) must be assembled into one reviewable record.

  • A launch review board, readiness gate, or accountable owner needs a structured

go / no-go / defer / exception recommendation.

  • A migration-dependent release needs evidence that every dependency is ready.
  • A low-risk documentation-only change needs proportional (not full-scale) readiness

review.

When not to use

  • Release pipeline mechanics — promotion, canary stages, CI/CD gate configuration,

versioning, and artifact management. Route to release-engineering. Release-engineering owns the pipeline gate; production-readiness owns the cross-domain evidence assembly and launch decision.

  • Incident response, SLO operations, on-call, and reliability engineering. Route to

site-reliability-engineering. SRE owns the live-service health boundary; production-readiness owns the pre-launch readiness boundary.

  • Security implementation, threat modeling, or penetration testing. Route to

secure-software-engineering or security-audit-methodology.

  • Data pipeline operations, schema migration engineering, or ETL/ELT design. Route

to data-engineering.

  • Test strategy, regression planning, and quality gates. Route to

qa-methodology.

  • Platform infrastructure provisioning or IDP design. Route to

platform-engineering.

  • Implementation planning, work breakdown, or delivery coordination. Route to

implementation-planning.

  • Spec authoring and SDD gate mechanics. Route to

spec-driven-development.

  • Verification verdicts against explicit criteria. Route to

verification-methodology.

  • API contract design and versioning policy. Route to

api-design-and-evolution.

  • Statistical analysis, experimental design, or causal inference. Route to

data-scientist.

  • End-to-end production lifecycle orchestration across all readiness dimensions.

Route to the production-excellence bundle (prose reference only — not yet landed). Production-excellence composes production-readiness, migration-engineering, resilience-and-recovery, capacity-and-cost-engineering, and incident-learning into a full production-operations lifecycle.

Risk classes and evidence scaling

The readiness evidence packet scales by risk class. A low-risk documentation release must not demand the same evidence as a user-facing service launch.

Risk classTrigger conditionsRequired evidence categoriesReview depth
LowDocs-only change, README update, non-functional config comment, internal tool that affects ≤1 team, content-only website change with no backendOwnership, user/business outcome (1-line statement), rollback (revert a commit)Self-review; lightweight record
StandardUser-facing feature, internal service with >1 consumer team, API addition, data-schema additive change, performance improvement with no SLO changeAll 11 categories (see checklist below); missing evidence requires explicit gap annotation with owner and datePeer review; full readiness record
HighCustomer-facing service launch, SLO-bearing change, auth/authz change, data migration with irreversibility, payment/billing path, compliance-scoped change, trust-boundary crossingAll 11 categories with evidence from a named source for every category; every gap requires a waiver with human approverFormal review; exception requires human approval annotation distinct from automated checks

Risk class is determined by the highest-risk dimension present, not by averaging or scoring. If any single dimension (e.g. data migration irreversibility, trust-boundary crossing) qualifies as High, the entire review is High.

Production evidence checklist

The 11 evidence categories form the minimum production evidence packet. Every category must have either a named source artifact or an explicit missing-evidence outcome.

#Evidence categorySource / evidenceGap / missing
1Ownershipsource: team name, on-call rotation, escalation path, or service catalog entrymissing: "no owner identified — deferred to <name>, due <date>"
2User/business outcomesource: success metric, OKR link, or product brief with measurable targetmissing: "outcome not defined — deferred to product owner, due <date>"
3Dependenciessource: dependency map, health-check results, or upstream SLO statusmissing: "dependency map incomplete — deferred to <name>, due <date>"
4SLOssource: SLO declaration, error-budget policy, or SLO dashboard linkmissing: "SLOs not declared — deferred to SRE, due <date>"
5Observabilitysource: dashboard link, alert rules, log/monitoring coverage, or Golden Signals reportmissing: "observability gap — deferred to <name>, due <date>"
6Supportsource: runbook, support playbook, escalation matrix, or support-handoff documentmissing: "no runbook — deferred to <name>, due <date>"
7Securitysource: security review record, threat-model summary, or security-acceptance sign-offmissing: "security review not completed — deferred to security team, due <date>"
8Datasource: data-classification record, retention/deletion policy, backup/restore test evidence, or migration test resultmissing: "data classification not available — deferred to data owner, due <date>"
9Rollbacksource: rollback runbook with rehearsal log, or revert-plan with recovery-time estimatemissing: "no rollback plan — deferred to <name>, due <date>"
10Capacitysource: capacity model, load-test report, quota/limit review, or cost projectionmissing: "capacity model not built — deferred to <name>, due <date>"
11Costsource: cost estimate, budget approval, or cost-attribution recordmissing: "cost estimate not available — deferred to finance owner, due <date>"

Structured fields (mandatory)

These four categories must appear as structured fields (checklist markers, table rows, or labeled form fields) — never only narrative prose.

  • [ ] Ownership — named owner or owning team with escalation path; recorded in the evidence checklist above.
  • [ ] Rollback — rollback plan or recovery path with rehearsal evidence; recorded in the evidence checklist above.
  • [ ] Support — runbook, support playbook, or handoff document; recorded in the evidence checklist above.
  • [ ] Observability — dashboards, alert rules, monitoring coverage; recorded in the evidence checklist above.

Launch decision outcomes

Every readiness review produces exactly one of four outcomes. The accountable owner is recorded with the decision.

OutcomeMeaningRequired evidenceAccountable owner
GoApproved — proceed to launchAll evidence categories for the risk class are satisfied with named sources; no unresolved gapsNamed launch approver (e.g., service owner, launch-review chair)
No-goBlocked — do not launchAt least one blocking gap exists with a named missing-evidence entry; the gap is material to the risk classNamed launch approver
DeferPostponed — re-review after conditions are metDeferred gaps are recorded with owner and due date; the deferral reason is explicitNamed launch approver + gap owners
ExceptionApproved with waiver — launch despite unresolved gapThe specific gap is named; a named human approver explicitly grants the exception; the exception annotation is distinct from automated checks and cannot be self-grantedNamed human approver (distinct from the automated review)

The exception outcome requires an explicit human-approval annotation — a named individual who approves the waiver. This approval is distinct from any automated check result and must be recorded separately (e.g., in an escalation record or exception log). The exception cannot be self-granted by the submitter.

Workflow

  1. Determine risk class. Map the change against the trigger-condition table above.

Select Low, Standard, or High.

  1. Collect evidence. For each of the 11 categories, record a named source or a

gap with owner and due date. Use the readiness record template.

  1. Assess completeness. Check that every category has an entry. A category with

a gap annotation is complete for the purpose of the review — the gap is visible.

  1. Produce a launch decision. Map the evidence against the risk-class requirements

and produce one of: Go, No-go, Defer, Exception.

  1. Record the decision. Commit the readiness record with the accountable owner

and, for exceptions, the explicit human-approval annotation.

No universal risk score

This skill does not produce a universal risk score, aggregate risk rating, or total risk number. Each evidence dimension stands on its own and is assessed against the risk-class requirements individually. If individual dimensions are evaluated separately (e.g., a security dimension assessment or a data dimension assessment), the rationale for any weighting is stated explicitly in that dimension's evidence entry — there is no single number that summarizes readiness.

File map

PathLoaded when
references/discovery-brief.mdUnderstanding the ownership boundaries with sibling production and engineering skills
references/readiness-record.mdFilling a readiness record template or reviewing a submitted record

Route-to table

When the need is...Route to
Release promotion, canary stages, CI/CD gates, versioningrelease-engineering
Live-service SLOs, incident response, on-call, error budgetssite-reliability-engineering
Security requirements, threat modeling, secure defaultssecure-software-engineering
Data pipelines, schema migration, ETL/ELTdata-engineering
Test strategy, regression planning, quality gatesqa-methodology
Platform infrastructure, IDP, service networkingplatform-engineering
Work breakdown, dependency mapping, rollout sequencingimplementation-planning
Spec authoring, acceptance criteria, SDD phase gatesspec-driven-development
Verification verdicts against explicit criteriaverification-methodology
API contract design and versioningapi-design-and-evolution
Statistical analysis, experimental designdata-scientist
Full production-operations lifecycleproduction-excellence bundle (prose reference — routes to a future bundle composing production-readiness, migration-engineering, resilience-and-recovery, capacity-and-cost-engineering, and incident-learning)
do mesmo repositório

Mais Skills

Todos os Skills