agentleFS
Arnie016Arnie016/fable-mythos/skills/skill-author/SKILL.md

skill-author

Skill0 starsChanged 4 months agoMITView on GitHub

Creates a new Claude Skill to fable-mythos standard - a tight SKILL.md plus eval cases and an example. Use when the user wants to turn a repeated workflow, a giant pasted prompt, or a recurring task into a reusable skill.

24 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • 0

    Common

    Most of the 19 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    One skill, one job: If the workflow splits into two jobs, split the skill or build the more repeated half.

    line 18
  2. Distinctive:

    Self-check: before delivering: description triggerable?

    line 21

“Common” is measured: kept by at least 6 in 10 of the 19 edited versions of this file we could read. “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
SKILL.md

Skill Author

Mission

Turn a repeated workflow into a skill package a cheaper model can execute. Prompts are temporary; skills compound — but only if they're small, testable, and honest about failure.

Step zero

Confirm the workflow is actually repeated. A one-off task gets an answer, not a skill — building a skill for something done once is how prompt libraries fill with corpses.

Workflow

  1. Description first. It does the triggering: name the task, the artifacts involved, and 3–4 phrasings a user would actually type. A skill that never fires is dead regardless of quality.
  2. Body under 60 lines: mission (2 lines) · workflow (≤7 numbered steps, each operational — a step that can't fail isn't a step) · rule precedence (explicit user requests beat style rules; say it) · output format (≤5 sections).
  3. Bulk goes to reference files. Voice, long examples, templates → examples.md in the skill folder, loaded on demand. If the user pasted a 2,000-word prompt, compress the logic into the skill and move the rest — never transcribe.
  4. One skill, one job. If the workflow splits into two jobs, split the skill or build the more repeated half.
  5. Write ≥4 evals (JSONL: id, type, input, expected, failure_to_prevent): at least one messy real-world input and one adversarial (the user asks for something the skill should resist).
  6. Write 1 example showing input → ideal output, annotated.
  7. Self-check before delivering: description triggerable? · every workflow step falsifiable? · body ≤60 lines? · adversarial eval present? · no "you are a world-class expert" preamble? · nothing the executor model can't follow?

Boundaries

Refuse skills engineered for dishonesty — auto-10/10 self-graders, fake-engagement generators, sycophancy by design. An evaluator that can't fail the work isn't a skill, it's a rubber stamp.

Output format

SKILL.md (complete)

evals/<name>.jsonl (complete)

examples/<name>-example.md (complete)

Handoff note

Where each file goes in the repo + the reminder: the skill gets good through the repair loop on the executor model, not through this authoring pass.

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.