agentleFS
Sign inStart free
williamcfranciswilliamcfrancis/humanize-code/AGENTS.md

humanize-code

AGENTS.md0 starsChanged 27 days agoMITView on GitHub

How to change humanize-code without breaking the skill, its package, or its evals. humanize-code is an agent skill written in Markdown, plus two helper scripts and an eval suite. There is no build step. Keep the skill portable. Do not write instructions that only work in one agent. Use Plain Language in prompts, docs, comments, and messages: lead with the point, use common words and active voice, keep sentences short, use one term per idea, and use must for requirements.…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repositoryOne other file it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep the skill portable. Do not write instructions that only work in one agent.

    line 17
  2. Very specific:

    Behavior first: Every before/after example in SKILL.md and the references must be behavior-identical.

    line 21
  3. Very specific:

    Version: Keep one version in SKILL.md metadata.version, the first CHANGELOG heading, and every manifest.

    line 25
  4. Very specific:

    Description: Every manifest uses the first sentence of the SKILL.md description.

    line 26
  5. Very specific:

    Length: Every word of SKILL.md is read on each use.

    line 27
  6. Distinctive:

    History: Add a CHANGELOG note for any behavior change or non-obvious fix.

    line 28

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Guide for agents

How to change humanize-code without breaking the skill, its package, or its evals.

What this repo contains

humanize-code is an agent skill written in Markdown, plus two helper scripts and an eval suite. There is no build step.

  • skills/humanize-code/SKILL.md is the prompt agents read and the source of truth. It explains why AI code looks the way it does, states Rule zero (behavior does not change), and lists numbered patterns grouped in sections, strongest first.
  • skills/humanize-code/references/ holds one file per language family. SKILL.md tells agents to read the matching file before any edit beyond comments.
  • skills/humanize-code/scripts/scan.py lists likely tells with pattern numbers. check_cosmetic.py proves an edit changed only comments and whitespace. Both use only the Python standard library.
  • skills/humanize-code/agents/openai.yaml holds the display name and default prompt for Codex.
  • evals/ holds four fixture projects with traps, suite.json describing their checks and traps, and run.py to prepare, grade, and trap-check runs.
  • .claude-plugin/, .cursor-plugin/, plugin.json (Agent Plugins 1.0.0), and gemini-extension.json are the package manifests.
  • scripts/validate-package.py checks that all of the above agree.

Keep the skill portable. Do not write instructions that only work in one agent.

Rules for changes

  • Behavior first. Every before/after example in SKILL.md and the references must be behavior-identical. If you are not sure, run it. A pattern whose fix can change behavior must say under what conditions it is safe, or be marked report only.
  • Patterns. Number them from 1 without gaps, grouped by section and strongest first. Fold a new tell into an existing pattern when one already implies it. If you add, remove, or renumber a pattern, update the README tables, the README section title, the § references in SKILL.md and the references, and the pattern numbers in scan.py.
  • References. A hazard goes in a reference file only if you ran code that shows it or an official doc states it. Cite the doc.
  • Evals. A new trap needs a test that fails when the trap is sprung. Run python evals/run.py check-traps before you commit; every trap must be caught. Never edit a fixture's snapshots by hand; regenerate them from the untouched fixture code.
  • Version. Keep one version in SKILL.md metadata.version, the first CHANGELOG heading, and every manifest. Do not add a top-level version field to SKILL.md.
  • Description. Every manifest uses the first sentence of the SKILL.md description.
  • Length. Every word of SKILL.md is read on each use. The validator caps it at 8,000 words; a change that adds words should earn them, and detail that only one language needs belongs in its reference file.
  • History. Add a CHANGELOG note for any behavior change or non-obvious fix.

Checks before publishing

python scripts/validate-package.py
python evals/run.py check-traps
npx skills add . --list
claude plugin validate .

Writing style

Use Plain Language in prompts, docs, comments, and messages: lead with the point, use common words and active voice, keep sentences short, use one term per idea, and use must for requirements. Keep exact identifiers, commands, paths, and behavior-bearing examples. Do not use em dashes. This repo is about removing AI tells, so its own text and code should not have them: run the skill on your change before you open a pull request.

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.

Check your setup, free