create-prd

// turbo-all

Create PRD / Architecture Design

Transform the ideation output into a production-grade architecture design document with explicit decisions on every axis.

Input: .memory/wiki/specs/ideation/ideation-index.md (must exist and be approved) Output: .memory/wiki/specs/YYYY-MM-DD-architecture-design.md + .memory/wiki/specs/ENGINEERING-STANDARDS.md + .memory/wiki/specs/data-placement-strategy.md

Depth standard: Every section must be specified to the point where a developer cannot misinterpret it. The specificity-standards rule (.agents/rules/specificity-standards.md) applies to every word of the output.

1. Read ideation output

Read .memory/wiki/specs/ideation/ideation-index.md.

If the file doesn't exist → STOP: tell the user to run /ideate first.

Completeness validation: If the file exists, verify it contains these required sections:

• ## Structural Classification
• ## Engagement Tier
• ## MoSCoW Summary (with at least 1 Must Have feature)

If any required section is missing → STOP: "ideation-index.md is incomplete — missing [section]. Run /ideate to complete ideation before proceeding."

Use the Structure Map in ideation-index.md to locate: constraints, personas, domain details (fractal tree), cross-cutting concerns, and role coverage.

Pay special attention to Project Surfaces in meta/constraints.md — it determines which tech stack axes apply.

Read ## Engagement Tier from ideation-index.md. If not set, ask the user and write immediately.

Read .agents/skills/prd-templates/references/engagement-tier-protocol.md — each shard reads the tier and adapts gates.

1.5. Deep ideation loading (scale-aware)

Read the ## Progress Summary table in ideation-index.md. Check these thresholds:

If ANY threshold is met → read .agents/skills/prd-templates/references/deep-ideation-loading-protocol.md and follow its full procedure. This produces a Domain Digest Table written to .memory/wiki/specs/architecture-draft.md as ## Ideation Digest.

If no threshold is met → skip this step. The standard ideation-index.md read from Step 1 is sufficient.

Why: For large ideation outputs (50+ features across 6+ domains), the Structure Map and MoSCoW Summary in ideation-index.md are insufficient for accurate architecture decisions. The digest reads every domain-level *-index.md to capture the full feature inventory, role coverage, and cross-domain dependencies without needing to load hundreds of leaf files.

2. Load skills

Read each skill SKILL.md listed in the frontmatter skills array.

Check .agents/skills/ for stack-specific skills. Read .agents/skills/find-skills/SKILL.md for community skill discovery.

Shard Overview

Progressive working artifact: .memory/wiki/specs/architecture-draft.md is written incrementally by shards 1–3 and read by shard 4 to compile the final dated architecture-design.md.

Orchestration

Create .memory/wiki/specs/architecture/prd-working/ directory if it does not exist. Read .agents/skills/prd-templates/references/workflow-checkpoint-protocol.md — all shards use this checkpoint system.

Step A — Run .agents/skills/create-prd-stack/SKILL.md

Step A.5 — Run .agents/skills/create-prd-design-system/SKILL.md

Step B — Run .agents/skills/create-prd-architecture/SKILL.md

Step C — Run .agents/skills/create-prd-security/SKILL.md

Step D — Run .agents/skills/create-prd-compile/SKILL.md

Shard failure recovery: If any shard (A through D) fails mid-execution:

1. Check .memory/wiki/specs/architecture/prd-working/workflow-state.md for the checkpoint — it shows exact step, item, and next action
2. Check .memory/wiki/specs/architecture-draft.md for the last completed section
3. Present the current state to the user: "Shard [N] failed at {current_item} (step {current_step}). Resume from failure point or restart the shard?"
4. Do NOT proceed to the next shard until the current shard completes cleanly

Step E — Quality gate

Self-check against Architecture rubric

Read .agents/skills/pipeline-rubrics/references/architecture-rubric.md and apply all 15 dimensions as the self-check.

Read .agents/skills/prd-templates/references/architecture-completeness-checklist.md and verify all items.

For any dimension that scores ⚠️ or ❌ → resolve it NOW. Loop back to the relevant shard.

Remediation loop guard: Track remediation attempts per dimension. After 3 failed attempts on the same dimension → STOP: present the dimension to the user as a known gap with context on what was tried. Include it in the review presentation as an unresolved item for user decision.

❌ STOP — do not call notify_user until all dimensions score ✅ and all checklist items pass.

Checkpoint cleanup: After quality gate passes, delete .memory/wiki/specs/architecture/prd-working/workflow-state.md — process tracking is no longer needed. Keep synthesis files (stack-synthesis.md, design-system-synthesis.md) as audit trail.

Depth audit

Re-read the entire architecture document and for EACH section ask: "Could a developer implement this without asking a single clarifying question?"

If no → identify what's missing, add the detail NOW, re-check.

Note: This is an internal self-check. For a rigorous independent audit, run /audit-ambiguity architecture after this workflow completes.

Step F — Request review and next steps

Use notify_user to present both documents with self-check results.

STOP — do NOT proceed until the user explicitly approves.

Next step

STOP — do NOT propose /decompose-architecture or any other pipeline workflow. The only valid next step is:

• /audit-ambiguity architecture — unconditionally mandatory. The self-check above cannot replace an independent audit. After the audit passes, the next step is /decompose-architecture.

Completion Gate (MANDATORY)

Before reporting completion to the user:

1. Memory check — Apply rule memory-capture. Write any patterns, decisions, or blockers from this workflow to .memory/wiki/. If nothing to write, confirm: "No new patterns/decisions/blockers."
2. Progress update — Update .memory/pipeline/progress/ tracking files if they exist.
3. Session log — Write session entry to .memory/pipeline/progress/sessions/.