Produce asset

Produce Asset — Render-Ready Prompt + Manifest Orchestrator

Production skill. Converts brief-graphic → render-ready prompts + asset manifest.

Core Question: "Can any downstream tool produce the right asset from this prompt without follow-up?"

Tool-agnostic — emits prompts + manifest, no API keys. No output variation (VERBATIM, tier C — _shared/options-selection.md): "more" = a manifest re-run.

Critical Gates — Read First

Non-negotiable. Full text: references/procedures/critical-gates.md [PROCEDURE].

1. Tool-agnostic. No image-gen APIs / Figma MCP. --publish / --api-render → BLOCKED — this stack emits render-ready prompts; it does not call render engines.
2. No hallucinated logos / brand marks. Missing logo on disk → solid-color placeholder slot, never a stand-in. Critic Gate 3.
3. Aspect ratio + safe zones are spec. Brief's aspect (1:1 / 4:5 / 9:16 / 16:9 / OOH) + safe zones verbatim into every prompt. No silent overrides.
4. Copy-to-render verbatim. Brief copy slots (headline, CTA, captions) instruct exact strings — no synonymizing, no "improving."

Before Starting

Required: the brief — docs/forsvn/artifacts/marketing/design-briefs/[slug].md (or a brief-landing-page asset-slot) — plus brand/BRAND.md + brand/DESIGN.md. Brief missing → NEEDS_CONTEXT (defer to brief-graphic); brand files missing → NEEDS_CONTEXT (defer to create-brand).

Pre-Dispatch (Cold Start)

Resolve from the brief + brand files first; ask once (bundled) only for what's missing:

1. Brief — brief-graphic slug or path. → routing only
2. Target platforms — defaults to the brief's target_platforms. → routing only
3. Render-mode hint — default export-mode. → routing only

Quality Gate

Critic (agents/critic-agent.md) verifies brief 04 spec compliance:

• Every brief slot has prompts/[slot-id].md — no orphans, no extras
• Aspect + safe zones verbatim per prompt
• Hallucinated-logo prohibition + placeholder rule active
• Copy-to-render verbatim; EXIF / aspect overrides absent
• Manifest verification checklist matches the brief's spec gates (aspect, safe zones, legibility, brand-mark + color fidelity)

Auto-fail: any invented brand mark or non-verbatim copy slot. FAIL → re-dispatch prompt-author with feedback (max 2 cycles, then BLOCKED). PASS twice with operator override → log via scripts/log-critic-override.ts per D8 contract.

Agents + Dispatch

2 agents, sequential: Prompt Author (agents/prompt-author-agent.md) per slot → Critic (agents/critic-agent.md) on assembled manifest. Single route (export-mode). Pseudocode, re-run triggers, chain position: references/procedures/dispatch-mechanics.md [PROCEDURE]. Shared 2-agent pattern: references/_shared/production-pattern.md § 4.

Artifact Contract

• Manifest: docs/forsvn/artifacts/marketing/produced-assets/[slug]/manifest.md
• Per-slot prompts: docs/forsvn/artifacts/marketing/produced-assets/[slug]/prompts/[slot-id].md
• Lifecycle: pipeline (regenerated; not canonical)
• Frontmatter: manifest — skill/version/date/status/slug/source_brief/target_platforms/slot_count/provenance; per-slot prompt — skill/version/date/slot_id/platform/aspect_ratio
• Provenance: input_artifacts = brief-graphic + brand/BRAND.md + brand/DESIGN.md. output_eval: null until eval scores rendered asset.
• Cross-stack: consumed by downstream renderers + future evaluate-content / evaluate-ad. Schema changes require atomic update with brief-graphic + brief-landing-page — never silently drift.

Full template + per-slot prompt schema: references/format-conventions.md [PROCEDURE].

Anti-Patterns

references/anti-patterns.md [ANTI-PATTERN] — 5 orchestrator (skip brief read, hallucinate logos, silent aspect overrides, copy synonymizing, render-mode misroute) + 4 cross-cutting marketing-stack rows.

Durable Rules (protected)

<!-- SLOW_UPDATE_START --> <!-- No pinned rules yet. Populate via the slow-update workflow (see references/slow-update-fence.md). Each pinned rule must (a) be procedural not instance-specific, (b) be earned from a regression or critic-flagged failure, (c) cite the artifact / decision record that justified pinning. --> <!-- SLOW_UPDATE_END -->

Completion Status

End with one status:

• DONE — manifest + all per-slot prompts written, critic passed, Critical Gates green
• DONE_WITH_CONCERNS — delivered, secondary issues flagged (e.g., thin reference-image suggestion)
• NEEDS_CONTEXT — brief-graphic OR brand files missing, or target platforms undefined
• BLOCKED — --publish / --api-render requested, OR critic FAILed twice

Execution

Offer the registry-gated fork (category image) — Brief-only: run the prompts, mark the manifest checklist (feeds evaluate-asset); Assisted/Direct: render via a verified engine. See _shared/execution-fork.md; record execution_mode. For Assisted/Direct multi-slot runs, batch-check the render surface first (_shared/capability-preflight.md). At brief-binding, bind the image target tool — inherit tool_targets or ask once per references/_shared/tool-target.md; prompt dialect tunes to it, tool-agnostic stays the default.

References

• references/procedures/{critical-gates, dispatch-mechanics}.md
• references/{format-conventions, anti-patterns}.md
• references/_shared/{production-pattern, eval-loop-spec, before-starting-check, manifest-spec, mode-resolver, anti-sycophancy, artifact-contract-template}.md
• references/_shared/execution-policy.md — session execution profile
• Upstream: skills/marketing/{brief-graphic, create-brand}/

Worked Example

A 2-slot brief → manifest + verbatim prompts, placeholder-logo discipline, critic PASS, Brief-only fork: references/examples/produce-asset-walkthrough.md [EXAMPLE].