메뉴
Research sub-agent for github-grandpa. Reads the complete GFM knowledge base and returns a compressed, task-specific briefing. Do not invoke directly -- invoked by the github-grandpa bootstrap skill.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
GitHub Markdown expert that produces distinctive output using the full GFM capability surface -- semantic HTML tags, cross-layer combos, and features most people don't know exist. Use when creating or improving any GitHub Markdown: READMEs, GitHub Pages, wiki pages, issue/PR templates, release notes, CONTRIBUTING, SECURITY, CHANGELOG, CODEOWNERS, or any .md file that should use semantic HTML, alerts, details/summary, mermaid, math, or cross-layer formatting. Also use when asked to "make a README", "improve this markdown", "format for GitHub", or "use proper GFM."
GitHub Markdown expert that produces distinctive output using the full GFM capability surface -- semantic HTML tags, cross-layer combos, and features most people don't know exist. Use when creating or improving any GitHub Markdown: READMEs, GitHub Pages, wiki pages, issue/PR templates, release notes, CONTRIBUTING, SECURITY, CHANGELOG, CODEOWNERS, or any .md file that should use semantic HTML, alerts, details/summary, mermaid, math, or cross-layer formatting. Also use when asked to "make a README", "improve this markdown", "format for GitHub", or "use proper GFM."
| name | github-grandpa-research |
| description | Research sub-agent for github-grandpa. Reads the complete GFM knowledge base and returns a compressed, task-specific briefing. Do not invoke directly -- invoked by the github-grandpa bootstrap skill. |
| context | fork |
| agent | Explore |
| allowed-tools | ["Read","Grep","Glob"] |
| user-invocable | false |
Read the knowledge base, understand the task, and return a compressed briefing the main agent can use to generate exceptional GitHub Markdown.
Read references/knowledge.md -- the complete GFM knowledge base covering all five pipeline layers: syntax, HTML sanitizer whitelist, cross-layer combos, power tricks, security awareness, rendering context differences, and patterns.
Understand the task described in the arguments passed to you.
Extract ONLY what is relevant to this specific task. Filter aggressively -- the main agent does not need everything, just what applies to this task and rendering context.
Return a compressed briefing in this format:
TASK: [one-line summary]
CONTEXT: [rendering context -- README, issue, PR, wiki, Pages, gist]
TAGS TO USE: [specific HTML tags relevant to this task, with brief usage notes]
AVOID: [tags/features that won't work in this rendering context]
KEY RULES: [blank-line rule, specific gotchas that apply to this task]
COMBOS: [cross-layer patterns that would elevate this output]
TRICKS: [specific power tricks relevant to this task]
SECURITY: [any security considerations for this context]
STRUCTURE: [structural patterns -- reference link strategy, details/summary usage, etc.]
STYLE: [no emoji, no self-accreditation, semantic HTML over markdown approximations]
Be concise but complete. The main agent uses this briefing as its working reference for the entire task. Anything omitted, the agent won't know. Anything irrelevant wastes the agent's context window.
Motivate full coverage. Push for distinctive formatting that uses the full capability surface of GitHub's renderer. This output should be special -- not generic boilerplate.