Using Superpowers

Using Superpowers Skill

Loads the dojo framework at session start: confirms the core disciplines are active, surfaces relevant lessons, and verifies clean state before any work begins. Does NOT replace the underlying skills — it sequences their loading and the session opening ritual.

When to Use

• At the start of every new session in a dojo-enabled repo.
• When the user says "use superpowers", "activate the dojo", or "start the framework".
• When another skill references the mandatory workflow.
• After resuming from a checkpoint or context handoff.
• NOT mid-session for already-active skills — re-loading wastes turns.

Prerequisites

• The dojo is installed in the repo (skills/, optional-skills/, tasks/, spec/).
• The view, edit, grep, and powershell Copilot tools.
• git available to inspect branch + working tree.
• tasks/lessons.md exists (created empty if first session).
• scripts/verify.sh (or scripts/run-checks.ps1) available for the session-start gate.

How to Run

textCopy code
1. Review `tasks/lessons.md` for relevant active lessons.
2. Open `tasks/todo.md` — is there work in progress?
3. Confirm `git status`: right branch, clean tree (or known WIP).
4. Acknowledge: "Dojo framework active. Core disciplines loaded."
5. Hand off to the trigger skill for the user's actual request.

Quick Reference

Procedure

Skill Activation Map

Always active (Core Disciplines):

• behavioral-foundation — The five non-negotiable prime directives (the charter)
• plan-before-code — Plan multi-step work before touching code
• verify-before-done — Prove your work with evidence
• self-improvement — Capture lessons, review at session start
• demand-elegance — Challenge hacky solutions
• autonomous-bug-fix — Full bug cycle, zero hand-holding
• safety-guardrails — Preflight check before destructive commands
• subagent-strategy — Delegate research and analysis

Workflow skills (activate in sequence):

• brainstorming → using-git-worktrees → executing-plans → requesting-code-review → finishing-a-development-branch

On-demand skills (load when needed):

• code-review — Reviewing PRs or diffs
• refactoring — Safe restructuring
• test-writing — Writing meaningful tests
• pr-workflow — Preparing merge-ready PRs
• debugging — Systematic complex debugging
• codebase-onboarding — Understanding unfamiliar repos
• dispatching-parallel-agents — Concurrent sub-agent work
• receiving-code-review — Processing review feedback
• skill-creator / writing-skills — Creating new skills

Session Start Ritual

Every session begins with:

1. Read memory/INDEX.md — What knowledge exists in the vault?
2. Query relevant context — bash scripts/memory-query.sh --type pattern --recent 5
3. Review tasks/lessons.md — Any un-promoted lessons?
4. Check tasks/todo.md — Is there work in progress?
5. Verify git status — Are we on the right branch? Clean state?
6. Acknowledge active skills — "Dojo framework active. Memory vault loaded. All skills live."

Session End Ritual

Every non-trivial session ends with:

1. Write session summary — memory/sessions/YYYY-MM-DD-summary.md
2. Promote patterns — Any lessons at 3+ occurrences? Move to memory/patterns/
3. Record decisions — Any architectural choices? Log in memory/decisions/
4. Capture preferences — User corrections about style/approach? Log in memory/preferences/
5. Rebuild the graph — bash scripts/link-index.sh
6. Update lessons metrics — tasks/lessons.md

Enforcement Rules

1. All skills are mandatory — No skipping steps in the workflow
2. No code without a plan (unless one-liner fix)
3. No merge without verification — scripts/verify.sh or equivalent
4. No "done" without proof — Tests, logs, diffs
5. Every correction becomes a lesson — tasks/lessons.md
6. Every session ends with a summary — memory/sessions/
7. Patterns are promoted, not buried — 3+ occurrences → memory/patterns/
8. Main branch is read-only — Work on feature branches/worktrees

Examples

Session Start:

Agent: "Dojo framework active. Memory vault loaded (4 decisions, 7 patterns, 3 preferences). Reviewed lessons.md (3 active lessons, 1 relevant to today's task). No work in progress. On main branch, clean state. Ready."

Skill Triggering:

User: "I want to add real-time notifications" Agent: brainstorming activates → asks Socratic questions → design approved → using-git-worktrees activates → creates feature branch → plan-before-code activates → writes plan to tasks/todo.md → executing-plans activates → works through tasks → requesting-code-review activates → self-reviews → finishing-a-development-branch activates → presents merge options

Pitfalls

• "I'll skip brainstorming, I know what to build" — You don't. Ask questions.
• "I'll just commit to main this once" — You won't "just" anything. Use a branch.
• "Tests take too long, I'll add them later" — Later never comes. Test now.
• "The plan is in memory only" — Write it in tasks/todo.md or it doesn't exist.
• "I already know everything" — Review tasks/lessons.md. You forgot something.

Verification

• memory/INDEX.md was read and active context surfaced.
• tasks/lessons.md reviewed and any relevant rules acknowledged.
• tasks/todo.md checked for work-in-progress before starting new work.
• git status confirms correct branch and clean (or known-WIP) tree.
• Acknowledgement message issued ("Dojo framework active…").
• All Core Discipline skills are loaded for the session.
• Session-end ritual scheduled (summary + pattern promotion + link rebuild).