outbound-enginelisted
Install: claude install-skill esmefong/gtm-skills
# Outbound Engine
Turn a qualified account with a live signal into outreach worth sending, in the brand's own voice,
with every load-bearing line traceable to its source. One conviction sits under every phase:
**the scarce asset in outbound is trust, and the engine protects it by refusing weak sends.** A
"do not send, here is why" verdict is a first-class deliverable, not a failure.
Inboxes are full of AI-personalized noise. What separates outreach that works is judgment (send
only when the signal genuinely predicts the buyer's situation) and provenance (every claim in the
message has a source). Both are enforced here, the second mechanically.
## What this consumes (the composed system)
| Input | From | What it provides |
|---|---|---|
| Account record with signal | `account-sourcing` (`accounts.json`) | Who, the dated why-now, the owner lane, the entry role |
| ICP | `icp-research` (`icp.md`) | Triggers, buying committee, decision criteria, how-to-speak rules |
| Voice directory | `brand-voice-guide` | Tonal pillars, writing standard, the Avoid list |
No account-sourcing run? The skill accepts a manual account + signal, but it applies the same
evidence standard: a signal without a date and source does not pass the gate. No voice directory?
It drafts in a plain professional register and says so; it never invents a brand voice.
## Hard boundaries
- **Role-level only.** Sequences target a role ("Head of AI / ML platform lead"), never a named
individual. Contact discove