kta-prd-pipeline
Decompose validated idea-brief.md into 4–8 per-feature PRDs with screens nested. Use after kta-idea-pipeline. Triggers — write PRD, decompose features, feature breakdown, screen list, MVP scope to PRD
SKILL.md.claude/skills/kta-prd-pipeline/SKILL.md
GitHubPlatform Tooling
- Use AskUserQuestion for blocking user choices or confirmations.
- Use WebFetch/WebSearch for current external docs or public web research.
- Use Task tool with subagent_type for delegation.
- Use Skill tool when chaining to another installed skill.
PRD Pipeline (Per-Feature)
Three phases — decompose → per-feature PRD → cross-feature review. Consumes
plans/{slug}/idea-brief.md, emits plans/{slug}/prd/ with one PRD per feature, a Mermaid
navigation graph, and a master INDEX.
When to Use
- After running
kta-idea-pipeline—idea-brief.mdvalidated andfeature_seeds:populated - "I have an idea brief, write the PRDs"
- "Break my MVP scope into feature specs with screen lists"
- "Generate per-feature PRDs from this idea"
- Re-running on existing
prd/to refresh after idea-brief changes
How the pipeline works
The pipeline pre-fills ~60% of PRD fields with sensible defaults (Settings always has account/notifications/theme/about; Auth always has email + 1 social provider; etc.). The student picks from options offered by the skill instead of writing fields from scratch. There are no modes, no levels, no toggles — this is the natural flow.
Workflow
- Phase 1 — Decompose — read
references/phase-1-decompose.md. Readidea-brief.md, validatefeature_seeds(5–8 items) present. AI proposes 4–8 features grouped by 5 user-journey buckets ( Acquisition / Activation / Core Loop / Retention / Settings). Student edits/approves. Hard cap 8 — skill refuses 9th. Monetization is OUT OF SCOPE in this phase — refuse paywall/IAP/subscription/billing features and tell the student to defer to a later phase. - Phase 2 — Per-Feature PRD — read
references/phase-2-feature-prd.md. Loop. Each feature gets one PRD usingreferences/feature-prd-template.md. Screens nested inside (NOT separate PRDs per screen). Saveplans/{slug}/prd/{NN}-{feature-slug}.md. - Phase 3 — Cross-Feature Review — read
references/phase-3-cross-review.md. Generate Mermaidnav-graph.mmd(screens as nodes, navigation as edges). Surface shared components and naming conflicts. WriteINDEX.md. - Phase 4 — Handoff (MANDATORY) — after
INDEX.mdis written, useAskUserQuestionto offer the next step. See "Handoff Prompt" section below.
Handoff Prompt (Phase 4)
Trigger condition (STRICT): fire ONLY after ALL feature PRDs from Phase 2 are written AND
INDEX.md + nav-graph.mmd from Phase 3 are written. Do NOT fire mid-loop after a single feature
PRD. Do NOT fire if any feature in the Phase 1 approved list is missing its {NN}-{feature-slug}.md
file. The handoff is a Phase 3 boundary — confirm all PRD files exist on disk before proceeding.
Immediately after writing plans/{slug}/prd/INDEX.md (last file of Phase 3), surface the absolute
path of the PRD directory in a one-line confirmation, then call AskUserQuestion:
header: "Next Step"
question: "PRDs written to {abs prd dir}. What next?"
multiSelect: false
options:
- label: "Run design", description: "Auto-run kta-design-spec with PRD dir path"
- label: "Pick feature", description: "Run kta-design-spec for ONE specific PRD only"
- label: "Exit to read", description: "Stop here so I can read PRDs first"
Branching:
- If user picks "Run design" → invoke
Skill toolwithskill: "kta-design-spec"andargscontaining the absolute path of the PRD directory (e.g.,args: "plans/{slug}/prd/"). Downstream skill takes over. - If user picks "Pick feature" → emit a second
AskUserQuestionwith one option per feature PRD file (label = feature slug ≤17 chars, description = 1-line feature summary from PRD). On selection, invokeSkill toolwithskill: "kta-design-spec"andargs= absolute path of the chosen single PRD file. - If user picks "Exit to read" → end the skill cleanly. Print the PRD dir path one more time. Do NOT proactively run any other skill.
This handoff is non-optional — every successful run of kta-prd-pipeline ends with this question.
If INDEX.md failed to write, skip the handoff and surface the error instead.
Inputs
plans/{slug}/idea-brief.md(required) — must includefeature_seeds:5–8 items- Optional — student overrides on proposed feature list
Outputs
plans/{slug}/prd/INDEX.md— master feature list with status + screen countsplans/{slug}/prd/nav-graph.mmd— Mermaid LR-flowchart of screen navigationplans/{slug}/prd/{NN}-{feature-slug}.md— one per feature (4–8 files)
Integration
- Previous stage —
kta-idea-pipeline(provides idea-brief.md) - Next stage —
kta-design-spec(consumes per-feature PRDs to emit per-screen design specs)
Constraints
- Hard 8-feature cap — Phase 1 refuses 9th feature; student must consolidate or defer to V2
- 5 journey buckets are FIXED — no ad-hoc buckets, no Monetization bucket in this phase
- Screens always nested inside feature PRDs — NEVER emit per-screen PRDs
- No project-stack-specific defaults — generic patterns only (no Supabase / DRE-KT mentions in feature catalog)
- Skill emits markdown only, no code
- Validates
feature_seedsexist in idea-brief before running — refuses with clear error otherwise - Default working directory —
plans/{slug}/ - Phase 4 handoff is mandatory AND fires ONLY after ALL feature PRDs + Phase 3 outputs (
INDEX.md,nav-graph.mmd) are on disk. Never trigger mid-loop. Offers "Run design" (auto-chains tokta-design-specwith PRD dir), "Pick feature" (single-PRD design run), or "Exit to read"