Home / skills/sharpen-what-is-working/SKILL.md

skills/sharpen-what-is-working/SKILL.md

name: Sharpen what is working description: >- Use when enough messages have gone out to read a pattern, when replies dry up, or when the user asks why nobody is answering.

Inputs: the outreach log, the drafts that went out, and every reply.

This needs at least 20 first touches to mean anything. Under that, say so and give the raw counts instead of a read.

1. Reply rate overall with the raw counts next to it, then split three ways: by fit score, by hook type (a post they wrote, a job listing, a launch, a funding round, a company reason with no personal hook), and by title.
2. Name the two patterns pulling ahead and the two dragging. Quote the actual subject lines and opening sentences from both ends rather than describing them.
3. Read the noes and the silences separately. A no with a reason is the most useful line in the log, so list the reasons and how often each came up.
4. Say what the pattern means for the list, not only for the message. A whole title or a whole segment answering nothing is a targeting problem, and no rewrite fixes that.
5. Propose three changes at most, each one testable on the next 20 sends. Get a yes before you change anything standing.
6. On a yes, update the voice notes, the hook preference, and the target titles in memory, and say exactly what you changed.

Never claim a difference from fewer than 10 sends on either side. Never drop a segment on one bad week. Never present a rate without the counts behind it.

Output: the read in chat with the counts, and the three proposed changes. Save the dated read under the prospecting folder.