Skip to content
SkillMixed2,291 tokens in use

Autoresearch

Optimization loops for landing pages, emails, ads and forms: generate variants, score them with five simulated experts, and evolve the winners before launch.

3,862 stars on the repository MIT Licenseericosiu/ai-marketing-skills · autoresearch
npx skills add ericosiu/ai-marketing-skills -s autoresearch
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing Autoresearch costs

25tokens idle
2,291tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Mixed

Script-ledMixedModel-led

Scripts do part of the work and the model's judgement does the rest.

This describes the skill's structure, how much of it is scripts. It does not say how well it works. Scripts: autoresearch.py.

Touches

shelldetectednetworkdetectedwrites filesdetectedcredentialsdetected

Declared tags come from the skill's frontmatter; detected ones from its scripts, its code blocks and the commands it names in inline code. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Good
  • Frontmatter8/15
  • Structure25/25
  • Examples12/15
  • Output15/15
  • Size13/15
  • Resources0/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Good
  • Runs shell commands−5

    does this on the machine, and the frontmatter does not say so

    SKILL.md, line 149
  • Writes files−8

    does this on the machine, and the frontmatter does not say so

    SKILL.md, line 175
  • Reaches the network−10

    does this on the machine, and the frontmatter does not say so

    autoresearch.py, line 25
  • Reads credentials−15

    does this on the machine, and the frontmatter does not say so

    autoresearch.py, line 60

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity88
  • Maintenance44
  • Community56
  • Maturity28
  • Momentum40

Autoresearch applies a Karpathy-style experiment loop to conversion copy that has no traffic yet. For each element of your content (a hero headline, a subject line, an ad description) it generates ten variants, has a panel of five simulated judges score them in one batched call, keeps the top three, and writes ten new variants that push the winning patterns further.

It stops when a variant reaches the score threshold (80 by default) or after three rounds, then cross-breeds the element winners into complete versions and scores those as whole units. Every run leaves three files: the optimized content, a full experiment log as JSON and a readable report that explains the winner.

What You Get

  • An optimized content file named after your input
  • An experiments JSON with every variant, every judge score and the top picks per round
  • A Markdown optimization report with the winner and its rationale
  • Two runner-up alternatives in case the winner does not feel right
  • Quality gates that say whether the result is marginal, shippable or strong

Common Use Cases

  • Tighten a hero headline, subheadline and CTA before a landing page goes live
  • Rework an email subject line, opening line and PS before a send
  • Test many ad headline and description options without spending on impressions
  • Improve the headline, bullets and button text of a lead form page

What Makes It Unique

It separates simulated pre-launch scoring from real traffic tests and says to run it first, then deploy, then validate. It batches every round into one API call, scores per content type with tailored dimensions such as would_open for email or trust for forms, and refuses to cross-breed until each element has its own winner. The fifth judge is meant to be replaced with your own founder voice.

Under The Hood

The skill ships a Python script that uses the Anthropic SDK and needs an ANTHROPIC_API_KEY in the environment. It reads your input file, calls the Claude API to generate variants, scores them in single batched calls, and writes the result files into a data folder (and the optimized file next to your output path). It also pauses a second between rounds to respect rate limits. Nothing is published or sent anywhere else, and the source file is overwritten only if you turn on auto_apply.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers