magnus919_agent-skills/bundles/production-excellence/SKILL.md

name, description, license, compatibility, metadata

name	description	license	compatibility	metadata
production-excellence	Route cross-domain production evidence (readiness, migration, recovery, capacity/cost, incident-learning) into a launch or operational decision — go, no-go, defer, exception, or escalation — with an accountable owner and a post-launch learning path. Compose production specialists without copying their runbooks. Do not use for incident command, release-pipeline mechanics, platform architecture, threat modeling, data-pipeline design, or any task owned end-to-end by a single specialist skill; do not use as a generic checklist detached from service ownership, risk, evidence, and verification.	MIT	Platform-agnostic methodology. No runtime dependencies, API keys, or external services required.	tags production-excellence, launch-decision, gate-model, evidence-packet, operational-handoff, readiness-review, cross-domain, production-acceptance, go-no-go, exception-escalation, post-launch-learning

Production Excellence

A thin composition bundle that assembles cross-domain production evidence into a defensible launch or operational decision. It owns the acceptance and handoff layer — the gate model that reads evidence from specialist skills and produces go / no-go / defer / exception / escalation outcomes with accountable owners. It does not own any specialist's runbook.

When to load this

Load when:

• A service or change is approaching a launch decision and evidence from multiple production domains must be assembled.
• You need a structured gate model (go/no-go/defer/exception/escalation) with conditions, evidence, and accountable owners.
• Cross-domain evidence (readiness, migration, recovery, capacity/cost, incident history) must be combined into one operational handoff record.
• A launch or change needs a post-launch learning path routed to incident-learning and product-lifecycle-learning.
• You are coordinating a production change across SRE, release, platform, security, data, and QA specialists and need a single acceptance contract.

When not to use

Do not load this bundle for:

• Incident command or SLO operations — those are owned by site-reliability-engineering.
• Release-pipeline mechanics, versioning, or deployment strategies — those are owned by release-engineering.
• Platform architecture or internal-developer-platform design — those are owned by platform-engineering.
• Threat modeling, security review procedure, or vulnerability assessment — those are owned by secure-software-engineering.
• Data-pipeline design, ETL, or storage architecture — those are owned by data-engineering.
• Test-strategy design, regression-suite management, or test-automation framework design — those are owned by qa-methodology.
• A generic checklist detached from service ownership, risk, evidence, and verification — every gate in this bundle requires a named service owner, assessed risk, verified evidence, and a declaration of the verification boundary. A bare checklist is never a valid outcome.

This bundle composes specialists. It never replaces them and never re-derives their methods. If the task is wholly within one specialist's domain, load that specialist directly.

Readiness routing table

The bundle routes each production concern to the specialist that owns it. The bundle itself owns only the acceptance and handoff layer — the cross-domain assembly and the gate decision.

Primary production-domain routes

Domain	Specialist skill	What the specialist owns	What the bundle adds
Production readiness	production-readiness	Risk-scaled evidence packet (11 categories), go/no-go/defer/exception launch decisions, accountable owners	Cross-domain assembly with migration, recovery, capacity/cost, and incident evidence; gate integration
Migration	migration-engineering	Expand/contract, compatibility windows, dual-running, backfills, reconciliation, cutover, recovery paths	Migration evidence as input to the gate model; handoff of migration verification to the operational record
Resilience and recovery	resilience-and-recovery	Failure modes, degradation choices, RTO/RPO, restore testing, DR, game days, failover, data integrity	Recovery evidence as a gate condition; exercise results feed the handoff record
Capacity and cost	capacity-and-cost-engineering	Demand/capacity/scaling/utilization models, unit-cost connection to SLO decisions, cost-constrained scenarios	Capacity/cost evidence as a gate condition; SLO/cost tradeoff decisions feed the gate model
Incident learning	incident-learning	Observed facts, causal hypotheses, contributing conditions, follow-up work mapping, verified closure	Pre-existing incident evidence as a gate condition; post-launch incidents routed back to incident-learning

Supporting specialist routes

Domain	Specialist skill	When routed
Reliability / SLOs	site-reliability-engineering	SLO/error-budget status required for gate entry; incident response for post-launch issues
Release mechanics	release-engineering	Release plan, rollout/rollback strategy required for gate entry
Platform	platform-engineering	Service-catalog entry, paved-road status for new services
Security	secure-software-engineering	Security review evidence for trust-boundary changes
Data	data-engineering	Data-quality and pipeline evidence for data-path changes
QA	qa-methodology	Verification evidence for all launches
Verification	verification-methodology	Boundary labeling and gap declaration for evidence assessment

Cross-domain entry evidence

Before the gate model runs, entry evidence must exist from every applicable domain. The bundle does not gather this evidence — it requires it. The complete evidence packet specification is in references/evidence-packet.md.

Summary:

• Every evidence domain (readiness, migration, recovery, capacity/cost, incident-learning) has a named source or an explicit gap with an owner and due date.
• The packet is usable for both new services and changes to existing systems — domains irrelevant to the change are explicitly marked "not applicable" with a reason.
• Missing evidence is never silently omitted. Every gap is recorded.

Gate and exception model

The gate model produces exactly one of five outcomes for every production change. Full definitions, conditions, and evidence requirements are in references/gates.md.

Outcome	Meaning	Key condition
Go	Authorized to proceed to production	All required evidence domains are sourced; no blocking gaps
No-go	Blocked; must not proceed	A required domain has a blocking gap, or an irreversible step has no verified recovery path
Defer	Postponed with explicit conditions	A non-blocking gap or dependency has a committed resolution date; re-evaluation is scheduled
Exception	Proceeds under an explicit waiver	A human authority (not the agent, not the service owner alone) approves a time-bounded, risk-bounded exception
Escalation	Decision escalated to a higher body	Irreconcilable gate conflict, trust-boundary security gap, cross-team authority gap, or regulatory boundary

Every outcome is anchored to service ownership, risk, evidence, and verification. No gate passes on a bare checklist. Each outcome names the accountable owner and records the evidence that supports it.

Operational handoff and post-launch learning

After a gate outcome is reached, the operational handoff record (references/handoff-record.md) is populated.

Post-launch learning paths

Launch outcomes and post-launch observations feed two learning routes:

1. Incident learning — post-launch incidents (SLO degradations, unexpected failures, capacity breaches) are routed to incident-learning. The handoff record provides the launch context; the incident-learning skill's verified-closure requirement ensures follow-up items are tracked to completion.

2. Lifecycle learning — expected outcomes recorded in the handoff (SLO targets, capacity assumptions, cost projections) are routed to product-lifecycle-learning for expected-vs-observed comparison at the handoff's review cadence. The lifecycle-learning skill's continue/improve/harvest/pivot/pause/retire decisions are informed by the gap between predicted and observed production behavior.

The handoff record is populated for every outcome — not only Go. No-go, Defer, Exception, and Escalation each produce a handoff record with the blocking condition, the follow-up path, and the accountable owner.

Loading and nested-skill behavior

This bundle is the discoverable entry point. It does not contain nested sub-skills under a skills/ directory. All routed skills are top-level catalog skills referenced via relative markdown links. Harnesses that support progressive disclosure will discover this bundle through its SKILL.md frontmatter and load the referenced specialists on trigger.

See AGENTS.md for agent-specific loading notes.

File map

Path	Loaded when
references/discovery-brief.md	Understanding the bundle's boundary against existing production and release skills
references/evidence-packet.md	Assembling cross-domain evidence for a production decision
references/gates.md	Running the gate model — go/no-go/defer/exception/escalation
references/handoff-record.md	Producing the operational handoff record and post-launch learning path
manifest.yaml	Machine-readable composition contract (schema v1): purpose, audience, stages, included skills, prerequisites, outputs, handoffs, conflicts, and eval suite; consumed by the lifecycle capability matrix