brief-landing-page

Landing Page Brief — Orchestrator

Coordinates evidence anchoring, hypothesis generation, architecture, per-section spec, asset slotting, conversion gating, and hand-off prompt composition into a single approved brief. Capability metadata (routes, prerequisites, load map, artifact contract) lives in routing.yaml. Agent table + 3 routes + 3 Approval Gates + critic verdict logic: references/agent-manifest.md. Methodology: references/playbook.md.

Core question: Could a designer (or Claude Design) build the right page from this brief without a single follow-up question?

Critical Gates — load first

• No brief without brand artifacts. Missing brand/BRAND.md or brand/DESIGN.md → return NEEDS_CONTEXT. Brief depends on tokens, voice rules, sacred elements. Design against the realized surface — the live site / shipped page / approved exploration / house CREATIVE-DIRECTION.md — not tokens alone; cite ≥1 or record the explicit token-only fallback (references/_shared/realized-surface-grounding.md).
• Conversion rubric is mandatory. Every section spec is gated by references/conversion-principles.md CP-01 → CP-14. Brand-good but conversion-bad = failure.
• Sacred elements are rails, not options. Logo geometry, primary palette anchor, tagline wording, signature treatments are "do not touch."
• Envelope: completeness is the gate. G6 checks each section for headline candidates + CTA + conversion checklist; 250-500 lines typical; hard FAIL ~200/~700.
• Don't inline the shared skill chain. Reference by section header; add page-specific overrides only.
• No placeholder testimonials, fake logos, or pretend numbers. Spec the slot ("Customer logo grid, 6 cells × 60px") and note "delete cell if not real" — never fabricate.

Inputs + Outputs

Full inputs table + output companions + per-slot artifacts: references/procedures/inputs-outputs.md. Hard-required: page route/campaign name, brand/BRAND.md, brand/DESIGN.md. Main artifact: docs/forsvn/artifacts/marketing/brief-landing-page/[slug]/brief.md. Always-emitted companion: handoff-implementation.md (universal coding-agent prompt, stack auto-detected, Asset Placeholder Rule verbatim). Optional companions per target_handoff: handoff-{claude-design,figma,designer}.md.

Quality Gate

Two critics run in parallel before delivery, both binary PASS/FAIL:

• Conversion critic scores against references/conversion-principles.md (CP-01 → CP-14). Full rubric in agents/conversion-critic-agent.md.
• Brand-voice critic scores sacred-element compliance, voice rules, surface language, token discipline, brief envelope (250-500 lines). Full rubric in agents/brand-voice-critic-agent.md.

Cycle 1/2 verdict matrix + per-FAIL routing rule (from critics' fix direction field, not hardcoded): references/agent-manifest.md § "Layer 5 critic verdict logic". DONE_WITH_CONCERNS is the floor — every concern visible.

Before Starting

Apply references/_shared/before-starting-check.md.

Hard gates (before any questioning)

1. Brand artifacts. brand/BRAND.md AND brand/DESIGN.md present. Missing → NEEDS_CONTEXT, recommend create-brand. Either >60 days stale → warn before proceeding.
2. Route classification. No existing page → Route A. Existing page or prior brief → Route B. Absence of analytics is NOT a blocker; label assumptions clearly.

Hard gates pass → proceed to Pre-Dispatch.

Pre-Dispatch + Mode

Run canonical Pre-Dispatch (references/_shared/pre-dispatch-protocol.md). Needed dimensions: page identity (route + name, always supplied), tier (conversion-primary vs -secondary; programmatic out of scope), hypothesis intent, goal (leads/signups/purchases/demos), route (A or B, resolved by hard gates).

Warm/Cold Start prompts + 4-question Cold Start + Write-back map + Project-Specific Workflows + Context-to-Pass + hard-block conditions: references/procedures/pre-dispatch.md.

Mode (references/_shared/mode-resolver.md): --fast collapses L1/1.5/2/3/3.5/4 to single-agent execution per layer, skips Layer 5 critic dispatch (critics noted as "skipped under --fast"). --fast does NOT skip Hard Gates, 3 Approval Gates, or Critical Gates 1-6. Lean default (no --fast needed): a ≤3-sentence single-scope ask with no prior artifacts auto-applies the per-layer single-agent collapse but KEEPS the Layer 5 critic; the 3 Approval Gates + Hard/Critical Gates always fire. Full multi-agent layers engage for broad, multi-section, or evidence-anchored (Route B) asks, or an upward override. Session execution profile (single-vs-multi): inherit per references/_shared/execution-policy.md.

Routing + Approval Gates

Three routes (Route A fresh LP · Route B existing-LP redesign with mandatory "What Changed from rev N-1" · Route C re-run with --rev=N) and three STOP gates (Gate 1 Hypothesis Selection after L1.5 · Gate 2 Architecture Approval after L2 · Gate 3 Final Brief Acceptance after L5 critic merge) — all three gates fire even under --fast. Full dispatch graphs, presentation format, response handling, spawn mechanics, single-agent fallback: references/agent-manifest.md + references/procedures/dispatch-mechanics.md.

Artifact Contract

• Path: docs/forsvn/artifacts/marketing/brief-landing-page/[slug]/brief.md (versioned re-runs: v[N]/brief.md for --rev=N).

• Companions: always handoff-implementation.md; optional handoff-{claude-design,figma,designer}.md per target_handoff. Per-slot asset-slots/{slot-id}.prompt.md written downstream by brief-graphic.

• Lifecycle: pipeline — versioned re-runs preserve prior versions.

• Frontmatter: 17 fields — see references/format-conventions.md and references/_shared/artifact-contract-template.md § provenance two-variants. Provenance is required so evaluate-landing-page can ground scoring on input_artifacts and _dev/eval/promote-to-experience.ts can walk output_eval.

• Body: 15 sections (Title block · Concerns · IMC Context · Hypothesis Approved · What Changed from rev N-1 · Page Architecture · Section-by-Section Spec · Asset Slots · What NOT to Do · Implementation Prompt · Hand-Off · Pre-flight Checklist · Skill Chain · Launch Plan + Results + Why This Works · Review Gate).

• Envelope: completeness-gated (G6); 250-500 lines typical.

• Cross-stack contract: consumed by human designers + coding agents + brief-graphic (per slot) + evaluate-landing-page cycles (when brief referenced from loop's strategy/). Schema changes require atomic update across upstream callers (plan-campaign) + downstream consumers (brief-graphic, coding agents, evaluate-landing-page).

Full artifact template byte-identical: references/format-conventions.md.

Chain Position

Previous: plan-campaign (optional — campaign context), create-brand (required) | Next: brief-graphic per asset slot (optional), then implementation (Claude Design / image-gen / human designer).

Re-run triggers: post-launch performance evidence, BRAND.md/DESIGN.md update, ICP refresh, traffic source pivot. Increment --rev=N.

Skill deference: post-launch CRO from real evidence → evaluate-landing-page inside run-pipeline. Single visual asset spec → brief-graphic. No brand → create-brand first. Headline variations only → write-copy. Non-LP page (blog, docs) → out of scope.

Anti-Patterns

Read references/anti-patterns.md before brief ships — 12 brief-specific + 4 cross-cutting + 3 design-handoff prompting (19 total). Most common: ignoring sacred elements (G1 sacred 4/4 auto-FAIL), brief too long (G6 envelope), coding-agent inventing asset URLs (G8b Asset Placeholder Rule), hero copy violating voice (G2 Forbidden Vocabulary single-hit FAIL).

Durable Rules (protected)

Completion Status

• DONE — both critics PASS (cycle 1 or 2), brief approved, artifacts written.
• DONE_WITH_CONCERNS — after 2 cycles, ≥1 critic still FAIL or mixed; concerns pinned at top of brief.md AND in frontmatter. User sees both reports at Gate 3 and ships consciously.
• BLOCKED — user rejected at a gate, or required input missing mid-flow.
• NEEDS_CONTEXT — BRAND.md or DESIGN.md missing; cannot proceed.

Execution

At brief-binding, bind the design target tool — inherit tool_targets or ask once per references/_shared/tool-target.md; tool-agnostic is the default. After Gate 3 PASS, offer the registry-gated fork (category design) — Brief-only: hand handoff-implementation.md to a coding agent / Claude Design / Figma / designer; Assisted/Direct: a verified engine builds it end-to-end (returns a URL → evaluate-landing-page), you approve at the gate. See execution-fork.md; record execution_mode. Re-invoke --rev=N after post-launch evidence triggers a redesign.

Worked Examples

Three end-to-end walkthroughs (Route A fresh LP, Route B evidence-anchored redesign, Route C --rev=N with mixed-critic verdict): references/examples.md.