AgentSkillsCN

formax-skill-capture

当我们希望将刚刚完成的 Formax 工作流(例如命令、叠加层、工具、钩子、权限、UI 一致性)转化为可复用的 Codex 技能,并将其存入 .codex/skills 目录下时,此技能尤为适用——包括脚手架搭建、护栏设置,以及最低限度的测试清单。

SKILL.md
--- frontmatter
name: formax-skill-capture
description: Use when we want to turn a just-finished Formax workflow (e.g. commands, overlays, tools, hooks, permissions, UI parity) into a reusable Codex Skill under .codex/skills, including scaffolding, guardrails, and the minimum test checklist.

Formax skill capture (create skills from our workflow)

What this is

This skill is a Formax-specific wrapper around skill authoring: it helps us reliably convert “we just figured out how to do X” into a repo-local skill under .codex/skills/<skill-name>/.

It complements the built-in skill-creator (system skill) by adding Formax conventions:

  • repo-local skills (checked into git)
  • “where to change what” file map
  • invariants (UI parity, model injection orthogonality, etc.)
  • minimum regression tests to lock behavior

When to use

Trigger this when the user says:

  • “把这个流程沉淀成 skill”
  • “以后别再忘了怎么改 /xxx”
  • “给我一个复用的修改清单/固定套路”

Minimal questions to ask (do not proceed until answered)

  1. Skill name (kebab-case), e.g. formax-slash-command-workflow
  2. 1–2 sentence description: when should Codex use it?
  3. Scope: which subsystem? (pick one)
    • slash commands / overlays
    • tools (spec/handler/presenter)
    • hooks
    • permissions/policy
    • stability/input routing
  4. Any explicit “don’t change” constraints? (UI copy/spacing/colors, contracts, etc.)

Output requirements

The resulting skill must include:

  • a short “Goal” section
  • “Where to change what” (file list)
  • “Patterns” (2–4 common patterns)
  • “Tests to update” (minimum regression set)
  • “Guardrails” (what must not regress)

Avoid long prose; prefer checklists.

Create a new skill (scaffold)

Use the scaffold script:

sh
bash .codex/skills/formax-skill-capture/scripts/new-skill.sh <skill-name> "<description>"

Then fill SKILL.md body with the content above.

Refactor policy for skills

When a skill is created from a workflow we already shipped:

  • Do not broaden scope (no extra features)
  • Do not rewrite UI (unless explicitly requested)
  • Write/extend tests first if behavior has regressed before