agentleFS
stevancrisstevancris/sre-ai-agent/skills/production-readiness-review/SKILL.md

production-readiness-review

Skill0 starsChanged 5 months agoMITView on GitHub

Review a service or feature for production readiness before launch. Use when a new service is about to go live, when an existing service is adding a major new capability, or when onboarding a service to SRE ownership. This is the primary shift-left skill — catching reliability, observability, and operational gaps before they become incidents. Trigger keywords: production readiness, PRR, launch readiness, ready to launch, ready for production, going to prod, new service, onboard service, pre-launch, launch review, is this ready, can we ship, pre-production checklist, launch checklist, go live checklist, service handover, SRE handoff.

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before loading context files, check if context/CONTEXT.md exists in the current directory.

    line 19
  2. Distinctive:

    Use the answers inline for this session.

    line 31
  3. Very specific:

    Read context/CONTEXT.md, context/company/tech-stack.md, and context/company/incident-severity.md.

    line 51
  4. Distinctive:

    Ask: "Is this service on the critical path for revenue or user authentication?" — if yes, default to Critical tier.

    line 68
  5. Distinctive:

    Document all deferred items with owners and dates

    line 247
  6. Distinctive:

    Add deferred items to team backlog

    line 249
  7. Distinctive:

    Do not add to on-call rotation until blockers are resolved

    line 254

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Production Readiness Review (PRR) Skill

Setup Check

Before loading context files, check if context/CONTEXT.md exists in the current directory.

If context/CONTEXT.md exists — read it and proceed normally.

If context/CONTEXT.md does not exist — this skill was installed standalone (e.g. via npx skills add). Ask the user these questions before proceeding:

  1. Role — junior-sre / senior-sre / sre-manager (shapes output depth and tone)
  2. Cloud provider — aws / gcp / azure / on-prem / hybrid
  3. Observability stack — e.g. Datadog, Prometheus+Grafana, New Relic
  4. Company name and primary services affected (if relevant to this task)

Use the answers inline for this session. For persistent setup across all skills, suggest:

pipx install sre-agent
sre-agent init

Purpose

The PRR is the SRE's primary tool for preventing incidents before they happen. It is a structured review that a service must pass before SRE accepts on-call ownership. A service that fails a PRR does not go to production — it goes back to the team with a remediation list.

This skill operationalizes the principle: reliability is designed in, not bolted on.


Instructions

Step 1: Load Context

Read context/CONTEXT.md, context/company/tech-stack.md, and context/company/incident-severity.md.

Identify:

  • Service name and owner team
  • Expected traffic at launch (RPS, DAU, data volume)
  • Launch date / deadline
  • Risk tier (see Step 2)

Step 2: Classify the Service Risk Tier

TierCriteriaPRR depth
CriticalRevenue path, auth, data storage, >50% of usersFull PRR — all sections mandatory
StandardUser-facing feature, significant trafficFull PRR — minor items can be deferred with plan
InternalInternal tooling, low traffic, no revenue pathLightweight PRR — abbreviated checklist

Ask: "Is this service on the critical path for revenue or user authentication?" — if yes, default to Critical tier.

Step 3: Run the PRR Checklist

Work through each section. For each item: ✅ Pass / ⚠️ Defer with plan / ❌ Blocker.

Blockers must be resolved before launch. Deferred items need an explicit owner and date.


Section 1: Observability

#RequirementStatus
1.1Four golden signals instrumented: latency, traffic, errors, saturation
1.2Structured logs with trace_id, service, level, request_id
1.3Distributed tracing spans at all service boundaries
1.4Dashboard exists showing golden signals at a glance
1.5Logs are queryable and retained per policy
1.6No sensitive data (PII, tokens, secrets) in logs

Blocker threshold: 1.1 and 1.2 must pass. Others are ⚠️ deferrable with 2-week plan.


Section 2: Alerting

#RequirementStatus
2.1SLI defined (what to measure)
2.2SLO defined (target value and window)
2.3Error budget policy defined
2.4Multi-window burn rate alerts configured (P1 and P2 thresholds)
2.5Every alert has a runbook link
2.6Alerts route to the correct on-call rotation
2.7Alert volume is reasonable (no alert storm expected at launch)

Blocker threshold: 2.1, 2.2, 2.4, 2.5 must pass. Service without SLO cannot be owned by SRE.


Section 3: Runbooks and Documentation

#RequirementStatus
3.1Runbook exists for each configured alert
3.2Runbooks include: triage steps, remediation, rollback, escalation path
3.3Service architecture diagram is current and accessible
3.4Dependency map documented (what this service calls, what calls this service)
3.5Data flow documented (what data enters and exits, where it is stored)
3.6Known failure modes documented with blast radius

Blocker threshold: 3.1 and 3.2 must pass. Invoke runbook-generator for any missing runbooks.


Section 4: Deployment and Rollback

#RequirementStatus
4.1Deployment is automated (no manual steps in the critical path)
4.2Canary or progressive rollout configured
4.3Rollback procedure is documented and tested
4.4Rollback takes < 10 minutes
4.5Feature flags available for high-risk functionality
4.6Database migrations are backwards-compatible
4.7Deployment does not require downtime (zero-downtime deploy)

Blocker threshold: 4.1, 4.3, 4.4 must pass. 4.7 required for Critical tier.


Section 5: Capacity and Scaling

#RequirementStatus
5.1Expected peak traffic estimated and documented
5.2Service load tested at 2x expected peak
5.3Autoscaling configured with appropriate min/max
5.4Resource requests and limits set on all pods
5.5Dependency capacity verified (DB, cache, downstream APIs can handle load)
5.6Rate limiting configured on public endpoints
5.7Circuit breakers configured for all downstream dependencies

Blocker threshold: 5.1, 5.3, 5.5 must pass. 5.2 required for Critical tier.


Section 6: Reliability Design

#RequirementStatus
6.1Service degrades gracefully when dependencies are unavailable
6.2Timeouts configured on all outbound calls
6.3Retry logic uses exponential backoff with jitter
6.4No single point of failure for Critical tier services
6.5Data loss scenarios identified and mitigated
6.6Service handles restart correctly (no state loss on crash)
6.7Health check endpoint returns correct status (not just 200 always)

Blocker threshold: 6.1, 6.2, 6.7 must pass.


Section 7: Security

#RequirementStatus
7.1Service account uses least-privilege IAM role
7.2No hardcoded secrets in code or IaC
7.3All secrets loaded from secrets manager at runtime
7.4Network exposure is minimal (not publicly accessible unless required)
7.5Input validation on all public endpoints
7.6Authentication and authorization implemented on all endpoints

Blocker threshold: 7.1, 7.2, 7.3, 7.6 must pass.


Section 8: On-Call Readiness

#RequirementStatus
8.1On-call rotation defined and engineers are trained on this service
8.2Engineers have been walked through runbooks in a dry run
8.3Escalation path defined (primary → secondary → manager)
8.4Service added to alert catalog
8.5Post-launch monitoring plan defined (who watches what, for how long)
8.6Launch communication sent to SRE team and stakeholders

Blocker threshold: 8.1, 8.3 must pass.


Step 4: Generate the PRR Report

Production Readiness Review: <Service Name>
Reviewed: YYYY-MM-DD | Tier: Critical / Standard / Internal
Reviewer: SRE Team

VERDICT: ✅ APPROVED / ⚠️ CONDITIONAL / ❌ NOT READY

Summary:
  Pass:     N items
  Deferred: N items (with plans)
  Blockers: N items

━━━ BLOCKERS (must fix before launch) ━━━
❌ 2.2 — No SLO defined. Without an SLO, there is no way to determine when this
         service is degraded or measure reliability over time.
         Owner: <team> | Fix: Define SLI + SLO using slo-monitoring skill | Due: YYYY-MM-DD

❌ 3.1 — No runbook for "HighErrorRate" alert.
         Owner: <team> | Fix: Run runbook-generator skill | Due: YYYY-MM-DD

━━━ DEFERRED ITEMS (launch with plan) ━━━
⚠️ 5.2 — Load test at 2x peak not completed.
         Owner: <team> | Plan: Run load test in staging within 1 week of launch | Due: YYYY-MM-DD

⚠️ 4.5 — Feature flags not implemented for new checkout flow.
         Owner: <team> | Plan: Wrap in LaunchDarkly flag before next major release | Due: YYYY-MM-DD

━━━ PASSED ━━━
✅ 1.1 Four golden signals instrumented
✅ 1.2 Structured logging with trace_id
✅ 2.4 Multi-window burn rate alerts configured
[... etc]

Next steps:
  1. Fix all blockers
  2. Re-review blocked sections
  3. Sign off and add service to SRE on-call rotation

Step 5: Conditional Approval

For CONDITIONAL verdict (deferred items only, no blockers):

  • Document all deferred items with owners and dates
  • Set a 30-day post-launch review checkpoint
  • Add deferred items to team backlog

For NOT READY verdict:

  • Share PRR report with service team and engineering manager
  • Schedule a remediation check-in (recommend 1 week)
  • Do not add to on-call rotation until blockers are resolved

Step 6: Post-Launch Monitoring Plan

For approved services, generate a launch monitoring plan:

Launch Monitoring Plan: <Service Name>

Launch date: YYYY-MM-DD HH:MM UTC
IC (on-call): <primary oncall>

Hour 0–1 (critical window):
  - Watch error rate (alert threshold: >1% for 2 min)
  - Watch P99 latency (alert threshold: >500ms for 2 min)
  - Watch saturation: CPU and memory on all pods
  - Check downstream dependencies: DB connection count, cache hit rate
  - Be ready to rollback: command pre-staged and tested

Hour 1–24:
  - Monitor error budget burn rate
  - Check for any new alert types firing
  - Review logs for unexpected patterns

Day 2–7:
  - Daily error budget review
  - Confirm deferred PRR items are on track
  - 1-week post-launch retrospective

Rollback trigger:
  - Error rate > 2% sustained for > 3 minutes → rollback immediately
  - Any data integrity concern → rollback immediately

Guidelines

  • A PRR is a review, not a rubber stamp. If something is wrong, say so clearly.
  • The owner of a blocker is always the service team, not SRE. SRE identifies; team fixes.
  • Deferred items are acceptable — but every deferral needs an owner and a date. Undated deferrals become permanent deferrals.
  • For Critical tier: involve the engineering manager in the PRR sign-off.
  • Persona (junior-sre): always ask a senior SRE or manager to co-sign a Critical tier PRR.
  • Persona (sre-manager): add a risk statement to the PRR report summarizing total risk exposure if the service launches with current gaps.
  • A PRR failure is not a blame event — it is the system working correctly.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.