NEVER create a skill with fewer than 25 assertions in eval.json
NEVER write a Guard Rail with multiple sentences -- single Do not X only
NEVER write Quality Gates as goals (try to, aim to) -- each must be binary pass/fail
description field must list specific trigger words/phrases for activation accuracy
Load docs/INSTRUCTION-ENGINEERING-PATTERNS.md before authoring any skill
Follow existing skills as templates (zonewise-scraper, honesty-protocol are canonical examples)
Guard Rail
Do not create a skill without running assertion count check: assert len(assertions) == 25 before claiming the eval.json is complete.
1---2name: skill-creator3description: Skill Creator4---56# Skill Creator78## Role9Own Platform Skills authoring as spec-compliant template generation, not freeform documentation writing.1011## Working Mode12Identify skill domain -> Apply SKILL.md template -> Generate 25 binary assertions -> Validate structure -> Output ready-to-commit files.1314## Focus Areas151. SKILL.md frontmatter -- name, description (trigger criteria), optional context/agent/allowed-tools162. Role declaration -- Own [domain] as [measurable framing], not [anti-pattern]173. Working Mode -- 4-step: Map -> Separate evidence -> Smallest intervention -> Validate184. Focus Areas -- 6-8 concrete boundaries/tradeoffs (not abstract concepts)195. Quality Gates -- verify/confirm/check/ensure/call_out (binary pass/fail, not aspirational)206. Output Format -- JSON schema or structured block showing expected output217. eval.json -- 25 binary assertions, pass_threshold=0.8, autoloop_compatible=true228. Guard Rail -- single Do not [failure mode] line (per INSTRUCTION-ENGINEERING-PATTERNS.md)2324## Quality Gates25- verify: SKILL.md has all required sections (Role, Working Mode, Focus Areas, Quality Gates, Output Format, Constraints, Guard Rail)26- confirm: eval.json has exactly 25 assertions with id 1-25, all type=binary27- check: description field contains specific trigger keywords (not generic phrases)28- ensure: Guard Rail is a single sentence starting with Do not29- call_out: Any Quality Gate that is aspirational rather than binary pass/fail3031## Output Format32Two files ready to commit:3334 .claude/skills/<skill-id>/SKILL.md (frontmatter + 7 sections)35 .claude/skills/<skill-id>/eval.json (25 binary assertions)3637Confirm structure with:38 ls .claude/skills/<skill-id>/39 python3 -c "import json; d=json.load(open('.claude/skills/<skill-id>/eval.json')); assert len(d['assertions'])==25"4041## Constraints42- NEVER create a skill with fewer than 25 assertions in eval.json43- NEVER write a Guard Rail with multiple sentences -- single Do not X only44- NEVER write Quality Gates as goals (try to, aim to) -- each must be binary pass/fail45- description field must list specific trigger words/phrases for activation accuracy46- Load docs/INSTRUCTION-ENGINEERING-PATTERNS.md before authoring any skill47- Follow existing skills as templates (zonewise-scraper, honesty-protocol are canonical examples)4849## Guard Rail50Do not create a skill without running assertion count check: assert len(assertions) == 25 before claiming the eval.json is complete.
Run npx skillmds@latest add breverdbidder/skill-creator in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Skill Creator It is listed under Coding & Dev Tools on SkillMD.
This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
breverdbidder (@breverdbidder) published this skill. Their other Agent Skills are listed on their SkillMD profile.
breverdbidder