agentleFS
neskuchnyneskuchny/content_factory/.claude/skills/perf-loop/SKILL.md

perf-loop

Skill1 starsChanged 2 days agoMITView on GitHub

Performance loop — how to read the metrics of your own publications relative to the channel norm, what to do with winners, anomalies and misses, how to update the hooks library and channel state, and how to carry winners over to other platforms. Use at the retro (stage 8) and when new scores arrive.

9 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use addskill, apply-firmware and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ryanportfolioSkill

    addskill

    Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.

    Also in 5 projects that use this

  • ryanportfolioSkill

    apply-firmware

    Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.

    Also in 5 projects that use this

  • +10 more from ryanportfolio

The file

Files
SKILL.md

Performance loop

Input (read, do not write — scripts write these)

  • analytics/scores.jsonl — the multiplier and label of each post relative to the channel × format norm.
  • analytics/baselines.json — the norms. analytics/registry.jsonl — which of our content units each post corresponds to (rubric, hook type, flow).

Why relative numbers

A million views against a channel norm of 2 million is a drop. A thousand → a million is an event. So compare a post with the norm of its own channel and format at the same age (72 h; for Telegram also 24 h), and measure channel growth by the trajectory of the norm itself.

Labels and actions

LabelMultiplierAction
ANOMALY≥ 3Break it into atoms: topic, pain point, hook (first 1–3 s or first line), length, pace, cover, CTA. Save it as a golden reference (style-memory). Propose 2–3 follow-ups and a carry-over to other platforms
WIN1.8–3Into targeting-map and hooks-library. The topic can be scaled up (sighting round → scale)
OK0.6–1.8Log only (published-log), so we do not repeat ourselves
MISS< 0.6Diagnose: if the hook is weak (low retention of the opening / few reactions despite good reach), propose a re-hook — new first 1–3 s and a new headline, same video, re-released on another platform or a week later. If the topic is weak, mark the topic

Small samples: with low_sample (norm based on < 8 posts) labels are provisional. The conclusion "this hook type works" requires ≥ 3 posts of that type on the platform. Account for the day of the week, news hooks and paid seeding (seeding distorts reach — flag it in the registry).

Update memory

  1. memory/channel-state.md — the norm now, a month and a quarter ago, the trend, the best post of the period.
  2. memory/hooks-library.yaml — for each hook type per platform: n, median multiplier.
  3. memory/targeting-map.md — topics, hooks, formats, lengths, covers: what works and what does not, with evidence.
  4. memory/platform-learnings.md — conclusions per platform and the results of carry-overs.

Cross-platform carry-over (once a week)

For every WIN/ANOMALY: the meaning stays, the wrapper changes to fit the platform (Telegram post → Reels with an infographic → carousel). If a topic has failed twice on a platform, write "do not carry X over to Y" in platform-learnings → "Filters".

Choosing hooks for next week

70 % — proven types (n ≥ 3, median ≥ 1.2 on this platform), 30 % — exploration. The hook type of each unit is recorded in the plan and in the registry; without it, learning is impossible.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.