Stats
Actions
Tags
Enforces concise responses, parallel tool execution, no redundant work, exploration tracking, and proactive context compression in every Claude Code session. Auto-applies at start.
How this skill is triggered — by the user, by Claude, or both
Slash command
/superpowers-optimized:token-efficiencyThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Core operating standard for all sessions. Apply permanently from activation.
Core operating standard for all sessions. Apply permanently from activation.
ls or findMaintain a mental index of repository exploration performed in this session. Before every Read, Grep, or Glob call, check this index and skip the call if the result is already known and the file has not been modified since.
| Situation | Action |
|---|---|
| File already read, not modified since | Do NOT re-read — use what you already know |
| File already read, but YOU edited it since | Re-read only the edited file |
| File already read, but another tool/agent may have changed it | Re-read — external changes invalidate your knowledge |
| Identical Grep/Glob pattern + same scope | Do NOT re-run — reuse the previous results |
| Similar but broader Grep/Glob pattern | Run the new search — it may surface new results |
| Context compression occurred (earlier turns disappeared) | Trust your remaining knowledge of file contents; only re-read if you genuinely cannot recall the content you need |
When Claude Code compresses earlier messages, you lose the raw tool output but retain your own reasoning and summaries. This does NOT mean you need to re-read everything — you still know what you learned. Only re-read when you need specific content (exact line numbers, precise syntax) that you can no longer recall.
Do not wait for context to auto-compress mid-task. Break proactively at logical seams — before compaction forces it in the middle of implementation where you need variable names, file paths, and discovered facts intact.
Break context here:
Break context by: invoking context-management to write state.md with discovered facts, then starting fresh with only state.md as input.
Why this matters: Auto-compaction at 95% context fill destroys the most recent content — exactly the variable names, discovered paths, and evidence gathered just before implementation. A proactive break at 50% preserves all of it.
Use context-management when cross-session persistence is needed:
Within a session, Claude Code handles context compression automatically — do not invoke context-management just because the session is long.
Before any multi-step task, identify all missing information and request it in a single message rather than asking across multiple turns.
Active silently for the entire session. No confirmation output.
npx claudepluginhub repozy/superpowers-optimizedTeaches the four operations of context engineering — Write, Select, Compress, Isolate — for managing token budgets, compaction strategies, and context partitioning to keep AI sessions sharp and efficient.
Applies token optimization rules to reduce context usage and response length. Always active, it minimizes file reads, avoids preamble, and batches tool calls.
Guides managing Claude Code context window with /compact, /clear commands, auto-compaction config, sub-agents, targeted reads, background tasks, and conversation flows for long sessions.