Routes all technical tasks to workflow skills before code edits, planning, debugging, or reviews. Enforces project setup gate for memory files like project-map.md.
npx claudepluginhub repozy/superpowers-optimizedThis skill uses the workspace's default tool permissions.
This skill MUST be invoked when any of the following occur:
Creates isolated Git worktrees for feature branches with prioritized directory selection, gitignore safety checks, auto project setup for Node/Python/Rust/Go, and baseline verification.
Executes implementation plans in current session by dispatching fresh subagents per independent task, with two-stage reviews: spec compliance then code quality.
Dispatches parallel agents to independently tackle 2+ tasks like separate test failures or subsystems without shared state or dependencies.
This skill MUST be invoked when any of the following occur:
Exception: Micro tasks (typo fix, single variable rename, 1-line config change) can skip the entry sequence entirely. Just do them.
If a user explicitly overrides a skill's behavior, follow the user. Skills are defaults, not mandates.
Before technical execution, select workflow skills explicitly and follow them.
Technical execution includes code edits, debugging, planning, review, test status claims, and branch integration actions.
Invoke token-efficiency at session start — applies to all sessions, always.
Fresh project gate — evaluate both conditions in order:
ls project-map.md 2>/dev/null — gate only fires if the file does not existIf both are true, pause before proceeding and tell the user exactly this:
Before I start: this directory has no memory files set up yet. That matters for how well I perform across sessions.
Without setup, every future session on this project starts from scratch:
- I re-explore the project structure even if I mapped it last session
- I re-read files I already understood
- I may re-propose approaches that were already tried and rejected
- I lose the "why" behind every decision the moment the session ends
A ~30-second setup changes that permanently:
git init— enables staleness tracking so I only re-read files that actually changed (creates.gitonly, nothing else)project-map.md— I read this at every future session start instead of re-exploring blindsession-log.md— auto-captures what was built and decided, so future sessions start with: "I see from last session that X was rejected because Y — building with that constraint already applied" instead of rediscovering itSet this up before we build, or start immediately?
Wait for the user's answer before continuing.
git init --quiet directly (do not ask again — the user just confirmed), then invoke context-management for map generation only. Return to step 3 when done. Note: context-snapshot.json will not be created in this session — the context-engine hook already ran at session start before git existed. It will be created on the next session start, provided the session is opened from this project's root directory. If no commits exist yet it will be mostly empty; it populates fully after the first commit.Classify the task as micro, lightweight, or full (see Complexity Classification below).
If resuming work from a prior session, read state.md if it exists. Use context-management to save state before ending a session with ongoing work.
If known-issues.md exists at the project root, read it to avoid rediscovering known error→solution mappings.
If project-map.md exists at the project root, read it to orient to the project structure without re-globbing or re-reading known files. Then check staleness:
git rev-parse HEAD and compare to the hash in the map header.
git diff --name-only <saved_hash> HEAD to find changed files. Re-read only those; everything else in the map is still valid. Then update the corresponding Key Files entries in project-map.md and refresh the git hash in the header to the current HEAD — this prevents the same files from triggering a staleness re-read on every future session.Follow the path for the classified complexity level.
Classify every task into one of three levels. Do not invoke a separate skill for this — decide inline.
If any of the following are true, classify as full immediately — do not evaluate the lightweight criteria:
When in doubt, classify as full. An unnecessary brainstorming session costs one extra round. Skipping brainstorming on a task that needed it ships a gap. The asymmetry is not equal — always err toward full.
All of these must be true:
Before classifying as lightweight: explicitly state in one sentence why each of the four criteria above is satisfied. Do not assume. If you cannot articulate any one of them clearly, classify as full.
Action: Go directly to implementation. Only gate: invoke verification-before-completion when done. Skip brainstorming, planning, worktrees, and parallel dispatch.
Exception: If a dedicated implementation skill exists for this specific task (check the Routing Guide), invoke it — lightweight skips workflow overhead, not implementation skills.
Anything that doesn't qualify as micro or lightweight.
Action: Follow the Routing Guide below for the full skill pipeline.
premise-check (run before brainstorming or planning)deliberation → brainstorming → writing-plansbrainstorming → writing-planssubagent-driven-developmentexecuting-plansusing-git-worktrees (run before implementation)systematic-debugging → test-driven-developmentverification-before-completionfinishing-a-development-branchrequesting-code-review / receiving-code-reviewdispatching-parallel-agentscontext-managementerror-recoveryfrontend-design standardsclaude-md-creator (applies at any complexity level — never implement directly)self-consistency-reasoner is invoked internally by systematic-debugging and verification-before-completion; do not invoke it directly. token-efficiency is always-on and invoked at step 1 of the Entry Sequence.For subagent handoffs, include only current task scope, constraints, evidence, and references to state.md when needed.
Avoid carrying forward long assistant reasoning chains unless they contain required artifacts.
When output feeds another agent/tool step, prefer JSON or YAML schemas defined by the active skill.
If a red flag appears, restart from Entry Sequence.