skill-template
Reference structure for authoring new skills in this repo. Use when writing or reviewing a SKILL.md for a new skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Reference structure for authoring new skills in this repo. Use when writing or reviewing a SKILL.md for a new skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | skill-template |
| description | Reference structure for authoring new skills in this repo. Use when writing or reviewing a SKILL.md for a new skill. |
| user-invocable | false |
Use this structure when authoring a new skill's SKILL.md. Adapt sections to fit — not every skill needs every section (e.g. drop "Prerequisites" if there are none).
# [Skill Name] — [Short 3-Letter Code]
[One sentence describing what this skill does and what capability it gives the agent.]
## When to use this skill
[1-2 sentences on when to invoke this skill vs. doing the task manually.]
## Input
`$ARGUMENTS` — [What the user passes, e.g. a file path, match number, log directory, subsystem name. "No arguments required" if none.]
## Prerequisites
[Anything that must exist/be true before running. Delete if none.]
## Steps
### 1. [First step]
[Specific instructions: exact tool, command, file path, or pattern.]
### 2. [Second step]
...
## Output
[What the agent should produce or report: a table, a summary with file paths, a modified file, a pass/fail verdict with evidence.]
## Verification
[Optional: how the agent should verify its own output before reporting, e.g. run the build, replay a log, compare metrics.]
## Example invocation
/[skill-name] [example arguments]
SKILL.md under 500 lines; move detailed reference material to separate files in the skill directory..claude/skills/<skill-name>/SKILL.md (project) or ~/.claude/skills/<skill-name>/SKILL.md (personal). The directory name becomes the /command name.Skill design principle (from FRC 254's "The Next Revolution: AI in FRC"): Skills give agents access to your robot's debugging tools. The minimum useful agent loop is: read logs → replay or simulate → compare results → iterate. Each skill should own one step of that loop. A well-scoped skill lets the agent accomplish a specific task reliably — a poorly scoped skill produces unpredictable results.