agentleFS
varunk130varunk130/claude-code-skills/skills/ai-marketing-growth/ai-launch-playbook/SKILL.md

ai-launch-playbook

Skill1 starsChanged 23 days agoMITView on GitHub

Produces a launch plan built for Artificial Intelligence (AI) products and features: a demo-led narrative, waitlist and early-access design with capacity gates, a channel plan for Product Hunt, Hacker News, LinkedIn, and developer communities, evaluation-backed proof, a day-0 to day-30 runbook, and pre-written failure-mode communications. Use when launching an AI product, agent, or AI feature; planning a waitlist or private beta for a model-backed product; preparing for a Product Hunt or Hacker News launch; or de-risking a launch where output quality, cost, or safety could go wrong in public.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    No plan for the viral screenshot of a wrong, biased, or unsafe output

    line 20
  2. Distinctive:

    No Tier 1 launch with any Red gate.

    line 50
  3. Distinctive:

    Treat launch-day signups as success before activation and week-4 retention are known

    line 148

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “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

AI Launch Playbook

AI launches fail in public in ways normal software launches do not - plan the demo, the capacity, and the bad screenshot before day 0.

What this skill is

A launch workflow for products whose core output is generated by models. It complements a general go-to-market plan by handling what is specific to AI: demos that must be reproducible, inference capacity and cost that scale with hype, evaluation results as the main proof, technical audiences who will try to break the product, and the near-certainty that someone will post a bad output. It produces a sequenced runbook with owners and pre-approved responses.

What it solves

  • Cherry-picked demos that users cannot reproduce, turning launch-day interest into "it does not work like the video"
  • Waitlists that collect emails but never convert, or let in more users than inference budget can serve
  • Launch posts on technical communities that read as marketing and get flagged or ignored
  • Quality claims with no benchmark or evaluation behind them
  • No plan for the viral screenshot of a wrong, biased, or unsafe output
  • Cost spikes from free-tier abuse in the first 72 hours

When to invoke

  • Planning a public launch or open beta of an AI product or agent
  • Shipping a major AI feature into an existing product with a broad user base
  • Designing a waitlist, invite system, or staged rollout for a model-backed product
  • Preparing a Product Hunt, Hacker News, or developer-community launch
  • Re-launching after a weak first launch or a quality incident

Phase 1: Launch tier and readiness gate

TierScopeTypical channelsReadiness bar
1New product or flagship agentAll channels, press, communityAll gates green
2Major featureOwned channels, community, customer emailQuality and cost gates green
3Incremental improvementChangelog, in-productQuality gate green

Readiness gates (score each Red / Amber / Green):

GateGreen means
QualityOffline evaluation pass rate meets target on a held-out set of realistic tasks
Reproducibility8 of 10 fresh attempts of the demo task succeed for a new user
CapacityRate limits and inference quota cover 3x the expected peak concurrent users
CostCost per active user at peak known; abuse limits on free usage set
SafetyRed-team pass on misuse, prompt injection, and sensitive-topic prompts
SupportOn-call rota, status page, and failure-mode responses approved

No Tier 1 launch with any Red gate.

Phase 2: Demo-led narrative

Build the story around one hero task the target user does weekly.

ElementRequirement
Hero taskReal, common, and painful; shown end to end in under 90 seconds
Before and afterTime, steps, or error rate without and with the product
Live versus recordedRecorded demos labeled; at least one unedited live run published
LimitsOne sentence on what it does not do well yet
Try-it pathThe viewer can reproduce the hero task within 5 minutes of signup

The limits sentence builds trust with technical audiences and preempts the "gotcha" thread.

Phase 3: Waitlist and early-access design

Weekly invites = min(inference budget / cost per active user per week, support capacity / tickets per user per week)
Design choiceRecommendation
Qualification2-3 questions (role, use case, data source) to prioritize Ideal Customer Profile (ICP) fit
PrioritizationICP fit first, then referrals, then sign-up order
CadenceFixed weekly waves so capacity and feedback are manageable
ActivationEvery invite lands on a guided first task, not an empty screen
ExpiryUnused invites expire after 7-14 days and are reissued
FeedbackIn-product thumbs and a short survey after the first successful task

Track waitlist-to-invite, invite-to-signup, and signup-to-first-good-output conversion per wave.

Phase 4: Channel plan

ChannelWhat worksWhat failsOwner
Product HuntClear hero task, short demo video, founder replying to every comment all dayVote-begging, vague taglines
Hacker News"Show HN" by a builder, technical detail, honest limits, free path to tryMarketing copy, gated signup, defensive replies
LinkedInFounder or leader post with the before-and-after and a customer quoteGeneric announcement graphics
Developer communitiesCode samples, Software Development Kit (SDK) or Application Programming Interface (API) quickstart, evaluation methodologyPosting the same link everywhere
Customers and waitlistPersonal email with the hero task and inviteMass blast with no next step
Press and analystsData story or customer outcome under embargoFeature list without proof

Stagger channels across launch week instead of one spike, so capacity and the team can absorb each wave.

Phase 5: Evaluation-backed proof

  • Publish the evaluation method: task set description, size, scoring rubric, and comparison baseline
  • Report results against the realistic alternative (manual process or a general-purpose assistant), not only other tools
  • Include failure categories and their frequency, not just the headline pass rate
  • Date every benchmark and name the model version behind it

Phase 6: Day-0 to day-30 runbook

WindowFocusKey actionsMetrics watched
Day -14 to -1ReadinessGates review, demo lock, channel drafts, failure responses approvedGate status
Day 0LaunchStaggered posts, all-hands comment coverage, live dashboardSignups, errors, latency, cost per hour
Day 1-3StabilizeFix top 3 failure patterns, tighten abuse limits, publish known issuesError rate, first-good-output rate
Day 4-14ConvertOnboarding fixes, follow-up content, case-study outreachActivation, day-7 retention
Day 15-30ProvePublish early outcome data, launch retrospective, decide next waveWeek-4 retention, cost per retained user

Phase 7: Failure-mode communications

Pre-write and pre-approve responses for:

ScenarioFirst response withinResponse pattern
Viral wrong or harmful output2 hoursAcknowledge, reproduce, state fix or mitigation, no blame on the user
Capacity outage or slowdown30 minutesStatus page, waitlist pause, timeline
Cost runaway from abuse1 hourRate limits tightened, free tier adjusted with notice
Data or privacy concern2 hoursClear statement of what data is used and retained, escalation to legal
Benchmark disputed24 hoursShare method and raw examples; correct publicly if wrong

Output

  • Launch tier and readiness-gate scorecard
  • Demo script with hero task, before-and-after, and limits statement
  • Waitlist design with capacity formula and wave schedule
  • Channel plan with owners and staggered timing
  • Evaluation summary ready for publication
  • Day-0 to day-30 runbook and pre-approved failure-mode responses

Operating rules

Always

  • Make the demo reproducible by a new user before announcing it
  • Size invites to inference budget and support capacity, not to hype
  • State known limits in the launch materials
  • Have a builder, not only marketing, present on technical channels
  • Watch cost per active user hourly during the first 72 hours

Never

  • Launch Tier 1 with a Red quality, capacity, or safety gate
  • Publish benchmarks without method, date, and model version
  • Gate a Hacker News launch behind a sales call or waitlist with no try-it path
  • Argue with users posting bad outputs; reproduce and respond with facts
  • Treat launch-day signups as success before activation and week-4 retention are known

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.