Firefly project planner

Firefly Project Planner

The interactive entry point for anyone planning a Firefly Services engagement. If a user knows roughly what they want to build but doesn't know which skills to read in which order, this skill walks them through 4 clarifying questions and outputs a structured project plan with specific skill recommendations.

This skill is paired with firefly-skills-catalog — the catalog is the reference, the planner is the guided experience.

When to use this skill

Use this skill when:

• The user is scoping a new Firefly Services project and wants a structured starting plan
• An FDE consultant is doing engagement intake at a new customer and needs a 90-day roadmap
• A customer engineering lead is asking "where do we start" with Firefly Services
• The user wants a curated subset of skills (not the full catalog) relevant to their specific use case
• The user mentions "scope", "plan", "design", "roadmap", "engagement intake", "where do I start", or "what should I read first"

Do NOT use this skill when:

• The user already knows the specific skill they need — invoke it directly
• The user wants to browse every skill — use firefly-skills-catalog
• The user is in the middle of an active task and needs immediate operational guidance — use the specific skill that matches the task

Step 1 — Ask the four scoping questions

Before recommending anything, ask the user these four questions. Do not skip them. Specificity in the answers determines the quality of the plan.

Q1 — Primary use case

"What are you trying to build? Pick the closest match:

1. Generative campaign assembly — produce many marketing assets from prompts + templates
2. Asset variation at scale — produce many derivatives of approved hero assets
3. Brand-aligned asset generation — generate assets that match a specific brand style
4. Photo / image processing pipeline — process customer-supplied images through Firefly + Photoshop + Lightroom
5. Video generation — short generative video clips
6. GenStudio extension — extend GenStudio for Performance Marketing with custom capabilities
7. Other / mixed — describe in one sentence"*

Q2 — Volume / scale

"What's the expected production volume per month, once running?

• Light (<1,000 API calls/month)
• Medium (1,000 – 100,000)
• Heavy (100K – 10M)
• Very heavy (>10M)
• Unknown / still scoping"*

Q3 — Audience / industry

"Which industry / customer type, and which best describes the team?

• Industry: media & entertainment / financial services / retail & ecommerce / healthcare / consumer products / technology / other
• Team maturity: net-new to Firefly Services / has experimented but no production / has production but wants to scale / has production at scale already"*

Q4 — Deployment target

"Where will this run?

• Customer's existing cloud (AWS / Azure / GCP) — which one?
• A SaaS the customer already operates
• Adobe Experience Manager / Adobe Experience Platform
• Standalone Adobe App Builder project
• Other / unsure"*

Step 2 — Match answers to planning tracks

The planning tracks are defined in config/planning-tracks.yml in this repo. Each track maps a real intent to an ordered sequence of skills. After receiving the user's answers to Step 1, match to one or more tracks:

User's primary use case (Q1)	Best-fit track(s)
Generative campaign assembly	build-campaign-assembly-pipeline, brand-compliance-stack
Asset variation at scale	variation-and-expansion-workflow, scale-to-batch-volume
Brand-aligned asset generation	train-and-deploy-custom-model, brand-compliance-stack
Photo / image processing pipeline	photoshop-composition-workflow, lightroom-batch-normalize
Video generation	video-generation-workflow
GenStudio extension	genstudio-extensibility
Other / mixed	Multiple — read the user's description carefully and pick 2-3 tracks

Then layer additional tracks based on Q2 (volume):

Volume (Q2)	Add these tracks
Heavy or Very heavy	Always add scale-to-batch-volume, cost-and-finops, troubleshoot-production-issue
Medium	Add troubleshoot-production-issue
Light	Add from-zero-bootstrap if team is new

And by Q3 (team maturity):

Maturity (Q3)	Add these tracks
Net-new to Firefly Services	Always add from-zero-bootstrap, jwt-migration (if legacy code)
Has production but wants to scale	Add scale-to-batch-volume, cost-and-finops
Has production at scale	Skip from-zero-bootstrap; assume current platform knowledge

Step 3 — Output the project plan

Render the plan as a structured response with three sections.

Section 1 — Recommended skills, in reading order

A consolidated, deduplicated reading list combining all matched tracks. Group by phase:

• Phase 1 — Foundation (first 1-2 weeks): bootstrapping, auth, storage refs, first working call
• Phase 2 — Build (weeks 2-6): the use-case-specific skills (generation, composition, processing)
• Phase 3 — Scale (weeks 6-12+): rate limits, batch pipeline, cost optimization, brand guardrails, troubleshooting

Section 2 — Architecture decisions to make in the first sprint

Three to five concrete decisions the team should make in week 1, based on the user's answers:

• Auth pattern: OAuth Server-to-Server (always — JWT is dead)
• Credential isolation: per-customer vs per-environment
• Storage backend: where source assets live, where outputs go (S3 / Azure Blob / GCS / Dropbox)
• Rate-limit posture: default quota vs request raise vs queue-fronted architecture
• Brand-guardrail layer: custom models, content filtering, human review, asset tagging — at what stage

Each decision points to the specific skill that covers it in detail.

Section 3 — Open questions for the customer

Things the team needs to verify with the customer before architecture is finalized:

• IMS org has Firefly Services entitlement provisioned (often a separate SKU from base Adobe products)
• Production rate-limit allowance negotiated with Adobe account team
• Storage bucket and pre-signed URL conventions agreed with customer InfoSec
• Custom model training data ownership and licensing terms
• Brand guidelines / approval workflow integration touchpoints

Step 4 — Save the plan

If the user is doing engagement intake, offer to save the plan as a markdown document in the customer's workspace. The output of this skill is meant to be a living document — the team revisits it as the project progresses.

Validate

A project plan is well-formed when:

1. The recommended skill list is deduplicated (no skill listed twice across tracks)
2. The skills are ordered by phase (foundation → build → scale)
3. Every recommendation traces back to a specific user answer (don't recommend firefly-video-model if Q1 wasn't video)
4. The architecture decisions are concrete enough to act on this week (no vague "consider...")
5. The open questions are customer-actionable (the customer can answer them; the FDE consultant shouldn't have to)

Chaining with other skills

After the planner runs, the recommended next steps are typically:

• firefly-skills-catalog — for the user to browse the full catalog if they want to go beyond the plan
• firefly-services-bootstrap — almost always the first action item for a new project
• The specific use-case skill identified by Q1

References

• config/planning-tracks.yml — the planning-track definitions this skill uses
• firefly-skills-catalog — full machine-generated index of every skill
• scripts/catalog/build-catalog.ts — the generator that keeps the catalog and planner in sync with the rest of the repo