learning-loop
Log posts and learn from real numbers using scoreboard.csv, learning.md and payouts.csv. Use after a post goes out, at 12/24/72 hour reads, for the daily or weekly review, or when the user asks what is working or what they earned.
Bodila51 · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add Bodila51/gtavi-claudecode --skill learning-loop技能说明
name: learning-loop description: Log posts and learn from real numbers using scoreboard.csv, learning.md and payouts.csv. Use after a post goes out, at 12/24/72 hour reads, for the daily or weekly review, or when the user asks what is working or what they earned.
Learning loop
Files in the page folder: scoreboard.csv (one row per post), learning.md (Wins, Losses, Current Rules), payouts.csv (real payouts only), log.md.
Rules
- Real numbers only. Ask the human to paste metrics or a screenshot. If a number is not available, leave it blank. Never guess.
- Log every post when it goes out: post_id, date, time, platform, pillar, format, hook, length_sec, credit, test_name.
- Reads at 12h, 24h and 72h fill views_12h, views_24h, views_72h, then likes, shares, replies, saves, followers_at_read.
- One test per day. Change one variable only (hook style, length, posting time, caption length, still vs clip). Name it in
test_name. - No verdict under 24 hours. Prefer the 72h read if the post is still moving.
payouts.csv: record only amounts the campaign or platform actually shows. Never project earnings.
Daily (evening)
Best and worst post, one lesson written as a rule. Append to learning.md. Pick tomorrow's one test.
Weekly
Top 3 and bottom 3 posts, median views by pillar, by time slot and by platform. Double down on what is above median. Rework or cut what stayed below median for a full week. Pick next week's tests. Update the customized playbook.md with what changed. Send the human a short summary.
Late-night reads: say what is still missing instead of filling it in.