Agent-Skills.md

Agent Skills: Gaia Product Process

Provides the money-gated lifecycle and coordination playbook for taking a consumer one-off / in-app-purchase product from idea to validated bet through launch and live-ops. Use it to classify a product request, run each of the 10 stages as fan-out -> adversarial review -> synthesis against one shared unit-economics model, and apply net-of-fee money gates with explicit loop-backs. Use when coordinating product discovery, monetization design, validation, soft-launch, or live-ops for non-SaaS consumer software.

UncategorizedID: frostaura/ai.toolkit.gaia/fa-product-process

Repository

frostaura/ai.toolkit.gaia
frostauraLicense: MIT
3026

Install this agent skill to your local

pnpm dlx add-skill https://github.com/frostaura/ai.toolkit.gaia/tree/HEAD/plugins/product/skills/fa-product-process

Skill Files

Browse the full folder contents for fa-product-process.

Download Skill

Loading file tree…

plugins/product/skills/fa-product-process/SKILL.md

Skill Metadata

Name
fa-product-process
Description
Provides the money-gated lifecycle and coordination playbook that takes a consumer one-off / in-app-purchase product from an idea to a validated bet and on through launch and live-ops. Use it by classifying the product request, running each of the 10 stages as fan-out -> adversarial review -> synthesis against one shared unit-economics model, and applying net-of-fee money gates with explicit loop-back edges. Use it when a consumer-product goal needs framing, discovery, monetization design, demand validation, soft-launch or live-ops coordination, when a stage needs review and a gate before it may advance, or when a gate has failed and the work must loop back to the right upstream stage. It never covers B2B seat-based SaaS, and it never runs software construction, testing or release gating - a validated bet is handed to the engineering plugin.

Gaia Product Process

Scope and when to use

Use this skill to route a consumer-product bet through Gaia's product-discovery team. Scope is one-off + IAP/IAG monetization; B2B seat-based SaaS is out of scope and auto-renewing passes are red-lined unless an explicit in-scope decision is recorded.

Use when:

  • a consumer-product goal needs framing, discovery, or monetization design
  • a stage needs fan-out, review, and a money gate before advancing
  • a gate fails and the work must loop back to the right upstream stage

Do not use when:

  • the product is B2B seat-based SaaS
  • the bet is already validated and only needs software construction (use the delivery roles)

Required inputs

  • product goal, constraints, monetization lane, and success bar
  • the current unit-economics model version (or authority to initialize v0)
  • prior-title priors when this is a sequel/follow-on (payer base, CRM, tooling)

Owned outputs

  • the run brief and per-gate pre-registered thresholds
  • per-stage synthesized artifacts with reviewer corrections retained
  • the Opportunity Thesis and the go / loop-back / kill trail

The money-gated lifecycle

Each stage exits only on a willingness-to-PAY gate (see fa-product-money-gate).

  1. Discovery — one buildable, monetizable bet; thesis + model v0.
  2. Ideation — one concept + committed monetization hypothesis; others killed/parked.
  3. Monetization design — economy, SKU ladder, ad stack; net LTV>CAC ≥ 3:1 + payback + cashflow.
  4. Validation — money-on-the-line; deflated WTP + ≥1 paid signal; measured CPI replaces placeholder.
  5. Vertical slice / MVP — lovable core loop + live sandbox IAP + instrumented funnel.
  6. Production & hardening — content, device/perf QA, anti-cheat, localization.
  7. Compliance & certification — ratings, loot-box/gacha law, privacy/ATT, tax, EULA.
  8. Soft-launch — real money + retention in cheap geos; multi-condition scale gate.
  9. Go-to-market & UA — repeatable creative+channel engine; positive blended CAC:LTV.
  10. Live-ops & portfolio — the revenue engine; sunset/maintenance/sequel decision compounding to stage 1.

Coordination pattern

  • The coordinator is the only coordinator; specialists never call each other.
  • Per stage: fan out the owning specialist(s) in parallel → run ≥2 independent reviewers → synthesize.
  • Fan-out is literal concurrency: independent specialists — and the ≥2 reviewers — launch as multiple Task invocations in a single message. A run that serializes parallel-safe work is a defect.
  • Await only what gates the next decision; long stage work (scans, experiments, soft-launch reads) runs as background subagents while independent branches progress.
  • Reviewers emit traceable CORRECTION from review: notes that stay in the artifact.
  • One shared, versioned unit-economics model is the passed state; specialists update it, never re-derive it.

Parallel execution shape

  • Within a stage, artifacts with no data dependency are produced concurrently (e.g. S1 trend scan, teardowns, and persona/JTBD work; S4 interview, fake-door, and landing-page tracks). Model reads are free; model WRITES serialize through its steward.
  • The money gate is a hard barrier: no work downstream of a gate starts before it passes. Across bets, stages may progress concurrently when they do not contend on the same model.
  • Concurrent artifact edits require disjoint file scopes per agent; overlapping scopes need isolated git worktrees merged deliberately. Two agents in one scope overwrite each other.
  • Register parallel branches as sibling todo items; gates, blockers, and proof are per-task, so branches complete independently without softening the completion contract.

Gate and loop-back rules

  • Pre-register success/kill thresholds BEFORE a stage's evidence is gathered.
  • Quote all revenue net-of-fee (commission + refunds + fraud + tax); require LTV:CAC ≥ 3:1, payback, and a cashflow check.
  • PASS → auto-advance. FAIL → route to the loop-back target. Uneconomic → kill.
  • Route failures to the true upstream owner (the loop-back edge), not to the most recent owner by habit.

Anti-patterns

  • do not advance on gross or optimistic economics
  • do not skip review or strip reviewer corrections
  • do not run stages in parallel across a causal gate
  • do not serialize independent branches; name what is parallel and what truly depends
  • do not move thresholds after data lands
  • do not relabel a subscription as a repeated one-off

Handoff and downstream impact

  • give each specialist the brief, the current model version, and the stage thresholds
  • give the synthesizer the specialist outputs plus reviewer corrections
  • hand a validated, built bet to the Gaia delivery roles for construction

Completion checklist

  • the stage artifact, gate result, and next owner are explicit
  • the model version and thresholds used are attached
  • loop-back targets are named when a gate fails

References