Skip to content

AGENT-SKILLS-STANDARD.md

Latest commit

 

History

History
95 lines (71 loc) · 3.72 KB

AGENT-SKILLS-STANDARD.md

File metadata and controls

95 lines (71 loc) · 3.72 KB
Raw

Agent Skills Standard : Rust-Claude-Skill-Package

Compliance checklist for the agentskills.org open standard and Anthropic Skill Authoring best practices. Based on Frappe v3.2.0 MIGRATION-AGENT-SKILLS-STANDARD pattern.

Mandatory frontmatter shape

Every SKILL.md MUST contain :

---
name: rust-{{CATEGORY}}-{{TOPIC}}
description: >
  Use when [trigger scenario].
  Prevents [anti-pattern].
  Covers [scope].
  Keywords: [technical terms + symptom-based + plain-language synonyms].
license: MIT
compatibility: "Designed for Claude Code. Requires rust 1.85+,edition-2024."
metadata:
  author: OpenAEC-Foundation
  version: "1.0"
---

Field rules

Field Rule Validator
name kebab-case, max 64 chars, matches dir name validate-frontmatter.js
description folded scalar >, starts "Use when...", contains Keywords: line, max 1024 chars validate-frontmatter.js
license matches LICENSE file in repo root validate-frontmatter.js
compatibility quoted string, mentions Tech + Versions validate-frontmatter.js
metadata.author "OpenAEC-Foundation" validate-frontmatter.js
metadata.version semver string "1.0" or "1.0.1" etc. validate-frontmatter.js

Keywords symptom-based rule

Per Frappe v3.1.1 expansion : Keywords: line MUST mix three types :

  1. Technical : exact API names, function names, error class names
  2. Symptom-based : plain-language descriptions of what user sees
    • Examples : "blank screen", "nothing shows", "app is slow", "can't login"
  3. Plain-language : beginner phrasings + synonyms
    • Examples : "how do I", "what is", "getting started", "set up X"

Good Keywords (Frappe frappe-syntax-serverscripts) :

Keywords: Server Script, frappe, sandbox, import, doc event, validate, on_submit, server script example, import not allowed, sandbox rules, which script type to use.

Bad Keywords (too thin) :

Keywords: serverscript, frappe.

Structural rules

Rule Why Validator
SKILL.md < 500 lines Overflow content goes to references/ validate-line-count.js
3 reference files required : methods.md, examples.md, anti-patterns.md Progressive disclosure validate-structure.js
NO README.md inside skill folder All docs in SKILL.md / references validate-structure.js
Section headings use : not em-dash Typographic standard validate-emdash.js
English-only content Skills are LLM instructions, not user docs validate-language.js
Deterministic language (ALWAYS / NEVER) No "you might consider", no hedging validate-language.js

Discovery manifests (Phase 6.5)

After skill creation, root must contain :

  1. package.json with agents.skills[] array (npm-agentskills standard)
  2. agents/openai.yaml with skills_directory: ./skills/source
  3. GitHub topic agentskills set via gh repo edit --add-topic agentskills

Generated by scripts/generate-manifest.js against the actual skill tree.

Reference URLs

Compliance verification

node /path/to/Skill-Package-Workflow-Template/scripts/validate-frontmatter.js .
node /path/to/Skill-Package-Workflow-Template/scripts/validate-line-count.js .
node /path/to/Skill-Package-Workflow-Template/scripts/validate-structure.js .
node /path/to/Skill-Package-Workflow-Template/scripts/validate-language.js .
node /path/to/Skill-Package-Workflow-Template/scripts/validate-emdash.js .

All must exit 0. Failure : root-cause-fix in skill, not workaround.