convert-package

# Convert Package - Full Sales-Asset Composite (the agency replacement)

> *"That's the Donahoe Method. Write like a human. Sell like you mean it. Close like you know it works."* - The Donahoe Method, summary

This skill is the suite-level orchestrator that takes a product brief and emits a complete, ship-ready sales asset built end-to-end through the Method. It calls 14-16 sub-skills across 8 phases, parallelizing where the DAG allows. The output is everything an agency would deliver - diagnostic, copy, page, integrity gate, visual specs, audit, client report - in one composite run.

**Note on `convert-transition`:** the transition library is consumed by `convert-chute` (which uses transitions as part of momentum engineering - open loops at section breaks, energy moves, etc.) rather than running as a standalone phase. When P3.5 invokes `convert-chute`, transitions are already integrated. There is no separate transition phase by design.

## When to Use

Trigger phrases: "convert package", "full sales asset", "agency replacement", "end to end sales page", "full method run", "donahoe full build", "build the whole thing", "everything for the launch", `/convert package <product>`.

Use when:
- The user has a product brief and wants the full deliverable, not just one primitive
- The user is replacing an agency engagement and wants the complete artifact bundle
- A net-new launch needs the asset stack assembled in one coordinated pass
- A re-launch requires rebuilding the page from the Method foundations up

Do NOT use for:
- A single primitive (use the specific sub-skill: `/convert open`, `/convert close`, etc.)
- Live-URL audits - use `/market landing <url>`
- Net-new offer construction - use `/funnel offer` first; package consumes the offer
- Funnel-level architecture (multi-asset campaigns) - use `/funnel build-campaign` (which can call this skill as one component)

## What This Composite Delivers

A complete asset bundle:

1. **Diagnostic** - The Four Questions answered + Market Temperature classification
2. **Method primitives** - The Open + The Three Locks + The Bullets + The Proof + The Cascade Close
3. **Sales page** - Full long-form direct-response page (copy + structure)
4. **Visual specs** - Above-fold hero spec + Scroll-rhythm map + Mobile compression + Proof-stack visual layouts
5. **Method audit** - 0-100 Donahoe Method score with prioritized fixes
6. **Client-ready report** - Aggregate of all of the above, formatted for handoff

Total artifacts: 12-14 files written to `out_path/<product-slug>/`.

## The Full DAG (8 phases, 14-16 sub-skill calls)

```
PHASE 1 - DIAGNOSTIC (sequential)
  └─ convert-four-questions <product>           [must run first; informs all others]
       └─ convert-temperature <product>         [depends on Four Questions output]

PHASE 2 - METHOD PRIMITIVES (parallel via delegate_task, max_concurrent_children: 5)
  ├─ convert-open <product>                     [depends on P1]
  ├─ convert-three-locks <product>              [depends on P1]
  ├─ convert-bullets <product>                  [depends on P1]
  ├─ convert-proof <product>                    [depends on P1]
  └─ convert-close <product>                    [depends on P1]

PHASE 2.5 - FINGERPRINT (sequential, after P2)
  └─ convert-fingerprint <product + P2 outputs> [develops only-you POV / contrarian voice that propagates through P3]

PHASE 3 - SALES PAGE ASSEMBLY (sequential)
  └─ convert-sales-page <product>               [composes P2 + P2.5 outputs into the page]

PHASE 3.5 - METHOD INTEGRITY GATE (sequential, three steps in order)
  ├─ convert-chute <P3 output>                  [engineers momentum: open loops, paragraph rhythm, transitions]
  ├─ convert-voice <P3 output>                  [applies Voice Rules - first person, contractions, no marketing-speak]
  └─ convert-bullshit-filter <P3 output>        [adversarial coaching pass - quotes 3 weakest lines, forces 3 rewrites each]

PHASE 4 - VISUAL SPECS (parallel via delegate_task, max_concurrent_children: 4)
  ├─ convert-above-fold <product>               [depends on P3 for headline + CTA]
  ├─ convert-scroll-rhythm <product>            [depends on P3 for section list]
  ├─ convert-mobile <product>                   [depends on P3 + P4.scroll-rhythm]
  └─ convert-proof-stack <product>              [depends on P2.proof + P3]

PHASE 5 - METHOD AUDIT (sequential, AFTER the integrity gate)
  └─ convert-audit <p3 output>                  [audits page AFTER P3.5 has had a chance to surface weaknesses; depends on P3 + P3.5 + P4]

PHASE 6 - CLIENT REPORT (sequential)
  └─ convert-report <all prior artifacts>        [depends on P1-P5; aggregates]
```

**Total sub-skill calls:** 14-16 (P1: 2 sequential; P2: 5 parallel; P2.5: 1 sequential; P3: 1 sequential; P3.5: 3 sequential; P4: 4 parallel; P5: 1 sequential; P6: 1 sequential).

**Why the integrity gate (P3.5) runs BEFORE the audit (P5):** the audit is an inspection - but the chute + voice + bullshit-filter steps actually FIX the draft. Running them after P3 (page assembly) and before P5 (audit) means the audit scores the post-fix version, not the raw page output. This avoids the "AI grades its own homework while still failing" problem identified in the Pass 1 cross-audit.

**The Bullshit Filter is a coaching pass, not a gate.** P3.5's filter call quotes the 3 weakest lines and forces 3 rewrites of each. The composite output includes those quotes + rewrites for the user to choose from. There is no "PASSED ✓" stamp anywhere - the user inspects the coaching artifact and decides whether to ship.

**Concurrency requirement:** This composite requires `delegation.max_concurrent_children >= 5` in the harness config. P2 fans out to 5 simultaneous children. P4 fans to 4 (within the same limit). If the harness is set lower than 5, P2 falls back to sequential - the asset still produces correctly, but slower (~3-5 minutes longer).

**Configuration check at runtime:**
```yaml
# In settings.json or pack-level config
dele