SuperAPI AI Directory

using-superpowers

General & other

View repository

Updated 10-15 23:33

About this skill

If a skill applies to your task, you have no choice. You must use it.

This is non-negotiable. You cannot escape through rationalization.

Rules

Invoke the relevant or requested skill before any response or action—including before clarifying questions, exploring the codebase, or viewing files. If after invoking it you find the skill does not fit the current situation, you do not need to use it.

Before entering EnterPlanMode: If you have not yet brainstormed, first invoke the brainstorming skill.

Then announce "Using [skill] to [purpose]", and strictly follow that skill. If it has a checklist, create a todo for each item.

Skill priority

When multiple skills apply, process skills take priority—they determine the approach, and then implementation skills (frontend design, etc.) handle execution. Brainstorming and systematic debugging are the most common process skills in Superpowers, but this rule applies to any process skill.

  • "Let's build X" → use brainstorming first, then implementation skills.
  • "Fix this bug" → use systematic-debugging first, then domain skills.

Red lines

These thoughts mean stop—you are rationalizing:

Thought Reality
"This is just a simple question" The question is the task. Check skills.
"I need to understand more context first" Skill check comes before clarifying questions.
"Let me explore the codebase first" The skill tells you how to explore. Check first.
"I can quickly check git/files" Files lack conversation context. Check skills.
"Let me gather information first" The skill tells you how to gather information.
"This doesn't need a formal skill" If a skill exists, use it.
"I remember this skill" Skills are updated iteratively. Read the current version.
"This doesn't count as a task" Action = task. Check skills.
"The skill is overkill" Simple things become complex. Use it.
"Let me just do this one thing first" Check before doing anything.
"Doing it this way feels efficient" Undisciplined action wastes time. Skills prevent this.
"I know what that means" Knowing the concept ≠ using the skill. Invoke it.

Platform adaptations

If your runtime environment is listed below, read the corresponding reference file for special instructions:

  • Claude Code:references/claude-code-tools.md
  • Codex:references/codex-tools.md
  • Pi:references/pi-tools.md
  • Antigravity:references/antigravity-tools.md
  • Copilot CLI:references/copilot-tools.md
  • Hermes Agent:references/hermes-tools.md
  • Qoder:references/qoder-tools.md
  • Muse:references/muse-tools.md

Gemini CLI users automatically get the references/gemini-tools.md tool mapping via GEMINI.md.

Chinese-specific skill routing

🇨🇳 This section is incremental content for superpowers-zh; upstream obra/superpowers does not have it. It explains when to use the chinese-* series of skills original to this fork. All other sections are translated section by section.

These 4 chinese-* areReference material, not a workflow — phrasing templates, formatting conventions, platform configuration differences. Theyare loaded only when the user explicitly invokes them, and should not be triggered automatically based on context: an extra formatting reference in the context won't make you write better, it will only crowd out what is actually needed.

The user explicitly says load the skill what's in it
/chinese-code-review chinese-code-review Chinese review phrasing templates, severity labels, responses to common anti-patterns in domestic teams
/chinese-git-workflow chinese-git-workflow Differences in SSH, credentials, and CI integration for Gitee / Coding.net / JiHu GitLab / CNB
/chinese-documentation chinese-documentation Spaces between Chinese and English, full-width/half-width punctuation, term preservation, Chinese copywriting typography guide conventions
/chinese-commit-conventions chinese-commit-conventions Conventional Commits Chinese adaptation, commitlint / husky Chinese templates

If the user hasn't named it, don't pull it in on your own. Conversely, if the user names it, follow it, and don't skip it just because it "looks like only a formatting issue".

They are used together with the translation skillin combination, not mutually exclusive: when the user asks for a Chinese review, it is requesting-code-review (process)plus chinese-code-review (style), not an either-or choice.

This "loaded only when explicitly invoked" is a deliberate decision, not an omission (see commit 392ff75). Changing it requires eval evidence — see PR #123 and this repo's CLAUDE.md "Skill changes require eval".

User instructions

User instructions (CLAUDE.md, AGENTS.md, GEMINI.md, etc., direct requests) take precedence over skills, and skills take precedence over default behavior. Only when your human partner explicitly tells you to skip may you skip a skill workflow or instruction.