| name | using-superpowers |
| description | Use when starting any conversation in Trae - establishes mandatory skill activation, Trae tool mapping, and workflow priority before any response or action |
If you were dispatched by Trae Task as a subagent for a specific task, skip this skill unless the task explicitly asks you to use it.
If you think there is even a 1% chance a skill might apply to what you are doing, you MUST use that Trae skill before responding or acting.
Trae may auto-load matching skills as context. If the required skill is not already visible in context, open or read .trae/skills/<skill>/SKILL.md before acting.
This is not negotiable. You cannot rationalize your way out of this.
Using Superpowers in Trae
Runtime Bootstrap
Trae should load this skill automatically through .trae/hooks.json on SessionStart. The same hook file also reinforces the workflow on UserPromptSubmit. If you are in a Trae session and this skill was not injected, open or read .trae/skills/using-superpowers/SKILL.md before doing any task work.
The runtime has four cooperating layers:
| Layer | File | Purpose |
|---|
| Hook | .trae/hooks.json | inject this skill at session start and reinforce each prompt |
| Agent | .trae/agents/*.md | provide named subagents for common Superpowers dispatch roles |
| Rule | .trae/rules/superpowers.md | keep mandatory triggers visible to the model |
| Skill | .trae/skills/*/SKILL.md | provide the workflow instructions |
| Rule reinforcement | .trae/rules/superpowers.md | keep the cross-session Superpowers contract without requiring Trae memory |
If these layers disagree, prefer direct user instructions first, then repository rules, then the current skill content.
Instruction Priority
- User's explicit instructions, project rules, and direct requests are highest priority.
- Superpowers skills define the required workflow and override casual default behavior.
- Default model habits are lowest priority.
If a user instruction conflicts with a skill, follow the user and state the conflict.
Trae Tool Mapping
When upstream Superpowers text mentions another harness, translate it to Trae:
| Upstream wording | In Trae |
|---|
Skill tool, superpowers:<name> | Trae auto-loaded skill context, or manually open/read .trae/skills/<name>/SKILL.md |
TodoWrite | Trae TodoWrite task list |
Task tool (general-purpose) | Trae Task subagent with the provided prompt template; use a matching .trae/agents named subagent when available |
Read, Write, Edit | Trae file tools |
Bash | Trae terminal/shell tool |
| local conversation memory scripts | not used in this Trae package |
Do not use legacy find-skills, skill-run, or remembering-conversations scripts in Trae. Skill discovery and activation are Trae-native and may appear as referenced context rather than an explicit tool call in the trajectory.
Subagent Selection
When delegating work, choose from the full available pool: Trae built-in subagents plus Superpowers named subagents in .trae/agents/. Pick the strongest agent for the user's current development need.
Strongest means the agent has the best fit across:
- Goal coverage: its description covers the user's actual requested work.
- Phase specialization: implementation, task review, code review, plan review, debugging, research, or verification.
- Workflow obligations: tests, evidence, read-only review, report format, or constraints.
- Scope control: it can work from the provided prompt and files without relying on hidden chat history.
Use these Superpowers agents when they match exactly:
superpowers-implementer for task-scoped implementation from a brief.
superpowers-task-reviewer for one-task review against a brief, report, and diff package.
superpowers-code-reviewer for broad code review after a completed feature or before merge.
superpowers-plan-reviewer for implementation-plan review before execution.
If a Trae built-in subagent covers the current domain or tool need better, use the built-in agent. If no subagent is clearly stronger or Trae Task is unavailable, work inline and state why.
Do not claim .trae/agents or a named Superpowers subagent is missing unless you have listed .trae/agents from the current target root in the same turn and confirmed the current working directory. If an agent was visible or invoked earlier, any later failure is a task/runtime failure, not proof that the agent directory disappeared.
The Rule
Use relevant or requested skills before any response, clarification, file read, shell command, implementation, or status claim.
Before entering plan mode or implementation planning: if you have not already brainstormed the work, use brainstorming first.
If you use a skill:
- Announce briefly: "I'm using
<skill> to <purpose>."
- If the skill has a checklist or multi-step process, create Trae
TodoWrite items for the steps.
- Follow the skill exactly unless the user explicitly overrides it.
If a skill contains prompt templates such as implementer-prompt.md or code-reviewer.md, load the template and pass its completed content to the strongest available Trae Task subagent:
superpowers-implementer for subagent-driven-development/implementer-prompt.md
superpowers-task-reviewer for subagent-driven-development/task-reviewer-prompt.md
superpowers-code-reviewer for requesting-code-review/code-reviewer.md
superpowers-plan-reviewer for writing-plans/plan-document-reviewer-prompt.md
Do not rely on session history as a substitute for the template or referenced files.
Skill Priority
Use process skills before implementation skills. Brainstorming and systematic-debugging are Superpowers' most common process skills, but the rule holds for all of them.
| Situation | First skill |
|---|
| New feature, build, rewrite, behavior change | brainstorming |
| Written spec or requirements need an implementation plan | writing-plans |
| Executing a plan with independent tasks | subagent-driven-development |
| Executing a plan inline or without subagents | executing-plans |
| Starting implementation work | test-driven-development |
| Bug, test failure, or unexpected behavior | systematic-debugging |
| Deep symptom with unclear original cause | root-cause-tracing |
| Flaky async tests or sleeps/timeouts | condition-based-waiting |
| Before claiming done/fixed/passing | verification-before-completion |
| Before merge, PR, or major handoff | requesting-code-review |
| Completing branch/worktree workflow | finishing-a-development-branch |
| Writing or updating skills | writing-skills |
| Stuck and unsure which problem-solving approach fits | when-stuck |
Local Problem-Solving Additions
These Trae package skills are shipped at the same install level as upstream skills:
| Situation | Skill |
|---|
| Force unrelated concepts together for breakthrough options | collision-zone-thinking |
| Flip assumptions to reveal hidden constraints | inversion-exercise |
| Recognize repeated patterns across domains | meta-pattern-recognition |
| Preserve two valid approaches without premature collapse | preserving-productive-tensions |
| Stress-test architecture or decisions at extreme scale | scale-game |
| Remove multiple components with one simplifying insight | simplification-cascades |
| Trace why an idea or technical choice evolved | tracing-knowledge-lineages |
Examples:
- "Let's build X" ->
brainstorming first, then implementation/domain skills.
- "Fix this bug" ->
systematic-debugging first, then domain skills.
Flattened Trae Skills
Upstream Superpowers v5 keeps some techniques as reference files inside parent skills. This Trae package also exposes the most important references as first-class skills so they can trigger directly:
condition-based-waiting
defense-in-depth
root-cause-tracing
testing-anti-patterns
testing-skills-with-subagents
Use the flat skill name when the scenario matches, even if the parent skill also links to the same material.
Red Flags
These thoughts mean stop and use the relevant skill:
| Thought | Reality |
|---|
| "This is just a simple question" | Questions are tasks. Check skills first. |
| "I need more context first" | Skill check comes before context gathering. |
| "Let me inspect files quickly" | Skills define how to inspect. |
| "I can check git/files quickly" | Files lack conversation context. Check skills first. |
| "Let me gather information first" | Skills tell you how to gather information. |
| "This doesn't need a formal skill" | If a skill exists, use it. |
| "I remember this skill" | Skills evolve. Use the current one. |
| "This doesn't count as a task" | Action equals task. Check skills first. |
| "The skill is overkill" | Simple things become complex. Use it. |
| "I'll just do this one thing first" | Check before doing anything. |
| "This feels productive" | Undisciplined action wastes time. Skills prevent this. |
| "I know what that means" | Knowing the concept is not the same as using the skill. Use it. |
| "I'll code first and test later" | Use test-driven-development first. |
| "The test failure is obvious" | Use systematic-debugging first. |
| "I manually verified it" | Use verification-before-completion before success claims. |
| "Task/general-purpose is a Claude thing" | In Trae, use the native Task tool with the template. |
Hook Health
The expected Trae hook set is:
SessionStart -> full using-superpowers bootstrap
UserPromptSubmit -> compact per-turn reminder
PreToolUse is intentionally not registered by default on Windows because hook stdin can be left open by some Trae host versions, stranding PowerShell processes.