메뉴
Enforces minimal context in agent configuration files like AGENTS.md, CLAUDE.md, GEMINI.md, and *.mdc to improve task success rates and reduce inference costs. Use when writing or modifying agent context files.
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
A simple example skill demonstrating the format.
Enforces ventilated prose and concise technical writing standards when the user is writing or editing standalone documentation files (e.g., Markdown, AsciiDoc). This skill should not be used for code documentation or docstrings.
| name | write-agent-context |
| description | Enforces minimal context in agent configuration files like AGENTS.md, CLAUDE.md, GEMINI.md, and *.mdc to improve task success rates and reduce inference costs. Use when writing or modifying agent context files. |
| metadata | {"purpose":"Provides context regarding the research at https://arxiv.org/abs/2602.11988"} |
A widespread practice in software development is to tailor coding agents to repositories using context files. Recent research indicates that providing excessive repository context tends to reduce task success rates. It also increases inference cost by over 20%. Both LLM-generated and developer-provided context files encourage broader exploration, such as more thorough testing and file traversal. While coding agents tend to respect their instructions, unnecessary requirements make tasks harder.
Therefore, human-written context files must describe only minimal requirements. Do not include generic documentation, tutorials, or non-critical context. Restrict these files to repository-specific constraints and critical architecture maps. Ensure all documentation formats are written with one sentence per line to improve diff readability. Documentation titles and headers must use sentence case.
