7 Claude Code Power Commands

Command	Status	Evidence
ultrathink	Already-have	ultracode_default.py hook forces max reasoning effort EVERY turn; the keyword overlaps. Confirm keyword still raises effort on demand.
/caveman	TO-ADOPT	No skill dir, not in the available-skills list. Genuinely new token-saver. Phase 2: verify source then install.
/insights	Already-have (equiv)	System Pulse (live /system-pulse) computes per-skill usage from session transcripts. Phase 3: decide merge vs keep-one.
/loop	Already-have	'loop' in the available-skills list; demonstrated this session via the cron heartbeat that re-fires the queue.
/schedule	Already-have	'schedule' in the available-skills list + scheduled-tasks MCP + 28 com.drwu launchd jobs already running.
/btw	TO-ADOPT	No skill dir. Non-interrupting side-question command. Genuinely new. Phase 2: trial during a long task.
/clear	Already-have	Built-in Claude Code CLI command to reset context between tasks. Adopt as a between-task habit (Phase 2).

Verdict: 5 of 7 already covered (ultrathink, /insights, /loop, /schedule, /clear). The 2 genuinely new, high-value items are /caveman (output-token reduction) and /btw (non-interrupting side question). Phase 2 installs and measures those; Phase 3 reconciles /insights with System Pulse and writes the standing rules.

Phase 2 finding (token-savers): The video gives no verified repo for /caveman, only "ask Claude to install it" — a blind remote install the plan itself bars (and a supply-chain risk). Deeper issue: /caveman cuts tokens by writing terse caveman-speak, which conflicts with the hard-gated Brandon reply format (full sentences, color-coded, anti-AI voice gate). So it cannot apply to user-facing replies. Recommendation: do NOT install /caveman for chat output; cost discipline is already handled by the DeepSeek-first ladder. /clear (built-in) is adopted as a between-task habit. /btw is a usage pattern to trial, not a discrete install.

Phase 3 decision (/insights vs System Pulse): KEEP System Pulse, do NOT run /insights as a second dashboard. System Pulse already exists, is live, and computes per-skill usage from the session transcripts; /insights would duplicate it. If a future /insights run surfaces a metric System Pulse lacks, fold that one metric into the System Pulse build. Standing command rules written to Brandon-Vault/40 Systems/Claude Code Command Rules.md. All 3 phases complete.

Source: ~/Projects/_outputs/claude-code-adoption/seven-claude-code-power-commands-plan.md, video rabGqnyd_Zw. Queue: q-2026-06-20-507e4c. Phase 1 of 3.