Outbound Engine

End-to-end cold outbound: ICP definition, expert panel copy scoring (recursive to 90+), infrastructure audit, capacity planning, and an implementation doc. This skill produces a strategy doc and ready-to-load copy for human review — it does not send. See references/workflow.md for the full procedure.

Core Rules

1. Determine mode first (existing Instantly account with API key, or starting from scratch) before any other step.
2. Run the three-phase workflow: Discovery & Audit → Expert Panel Recursive Scoring → Deliverables.
3. Expert panel scoring target is 90/100 — non-negotiable. Iterate until reached.
4. Show every scoring round in the final doc — the iteration trail is part of the value.
5. Never send or push to the platform autonomously. This skill's output is a doc for human review. Writes/sends happen only after explicit human approval — see the Sending gate below.
6. Use capacity math to set realistic volume and pipeline projections.

Operating Modes

Optimize existing — there is a live account to learn from. First questions: (1) Do you have an Instantly API key? (2) Which campaigns are underperforming, and on what metric? (3) Is there historical data to pull? With an API key, run scripts/instantly-audit.py to pull campaigns, account inventory, and warmup scores before touching copy.

Start from scratch — no campaigns yet. First questions: (1) Who is the ICP (titles, industries, company size)? (2) What do you sell, in one jargon-free sentence, and what is the primary offer (audit, trial, demo, call)? (3) What proof points are real and linkable? Use assets/icp-template.md to collect the ICP before writing copy.

The Three Phases (summary)

1. Discovery & Audit — Determine mode. If audit mode, run the Instantly audit and flag any account with warmup score <80 or <14 days as NOT ready. Define the ICP, capture business context (offer, real URLs, proof points), and confirm the expert-panel roster. Detail in references/workflow.md; ICP fields in assets/icp-template.md.
2. Expert Panel Recursive Scoring — Draft the sequence, then score it with the 10-expert panel (references/expert-panel.md), each scoring through their own lens. Each round emits a per-panelist score table, an aggregate, the top weaknesses, the edits made, and the revised copy. Score honestly — no padding to 90. Below 90: fix the top 3 weaknesses and run another round. At ≥90: finalize. Copy must follow references/copy-rules.md (and references/instantly-rules.md for Instantly variable/deliverability rules).
3. Deliverables — Produce the strategy doc: brutal-truth analysis, ICP summary, infrastructure status, every scoring round, final copy for every step, implementation plan, capacity math, weekly metric targets, and STOP/START lists. Full doc structure in references/workflow.md.

What the 90/100 Gate Means

90 is an operational gate, not a vibe. It means the aggregate average across all panelists hit 90 in a single round with no score padded to get there. Below 90, the copy is not done — identify the lowest-scoring lenses, address those specific weaknesses, and re-run. The point of the recursive trail is that each round names a concrete weakness and the edit that fixed it; ship the trail, not just the final copy.

Sending Gate (resolves Rule 5)

scripts/cold-outbound-sender.py and the write paths in scripts/instantly-audit.py exist but are gated. The default deliverable is the strategy doc. Do not invoke the sender or perform any platform write autonomously. Invoke cold-outbound-sender.py only when the human has explicitly confirmed the strategy doc is reviewed and approved AND explicitly asks to send. Always run --dry-run first and show the result before a live send. Never treat "build the campaign" as authorization to send.

Scope Note

The core skill is the three-phase design workflow above. Three scripts ship in this directory but sit outside that workflow and should not run unprompted: scripts/lead-pipeline.py (Apollo→LeadMagic→Instantly sourcing), scripts/competitive-monitor.py (competitor tracking), and scripts/cross-signal-detector.py (multi-source signal detection). Use them only on an explicit, specific request for that function.

Reference Files

Related skills

• content-ops — expert-panel quality gate; the recursive 90+ scoring of sequence copy is a content-ops review pass
• marketing-shaper — produces the scoped outbound brief (ICP, goal, structure) this skill executes from