| name | claude-code-power |
| description | Reference catalog of lesser-known, power-user Claude Code features — autonomous/unattended operation, channels and remote control, multi-agent orchestration, the full hook-event set, context/session surgery, obscure settings/env-vars/CLI-flags, and advanced skill authoring. Use when the user asks "what can Claude Code do", "is there a Claude Code feature for X", "how do I run Claude unattended / on a schedule / from my phone", "what hook events exist", "what obscure settings or env vars are there", "how do I configure/launch claude for Y", or wants to discover advanced harness capabilities beyond the basics. |
Claude Code Power Features
A curated reference of the high-leverage, non-obvious capabilities of the Claude Code harness. The basics (/clear, /compact, CLAUDE.md, plain subagents, plain MCP) are assumed and not covered here — this is the rest.
How to use this skill: this file is a router. Read only the reference file that matches the intent at hand — each is self-contained, so loading one does not require the others. Do not read all of them; pick by the table below.
Provenance and accuracy: sourced from code.claude.com/docs. Features list a minimum CLI version where one applies (e.g. "requires v2.1.81+"); check claude --version before relying on a gated feature. This is a snapshot, not the live docs — when a detail is load-bearing and the answer must be current, confirm against the source page. A handful of items are marked (verify) because the docs were internally inconsistent or silent on the exact mechanism; treat those as leads, not facts.
Routing table
| Read this file | When the goal is |
|---|
reference/autonomous.md | Keep Claude working unattended: /goal, /loop, scheduled tasks / cron, cloud routines + /fire webhooks, auto mode |
reference/remote-and-channels.md | Operate a session from elsewhere: channels (push events in), phone permission relay, remote control, deep links, mobile push |
reference/multi-agent.md | Split work across agents: advanced subagent frontmatter, agent teams, dynamic workflows, forks, worktrees & isolation |
reference/hooks.md | Intercept/automate on events: the full hook-event set (incl. the obscure ones) and hook mechanisms (prompt/agent handlers, if, args, terminalSequence) |
reference/context-and-session.md | Manage context & history: /btw, scoped rewind/summarize, transcript viewer, compaction steering, session forking, --from-pr |
reference/config-cli-env.md | Tune or launch the harness: obscure settings keys, behavior-changing env vars, CLI flags/subcommands, --safe-mode/--bare, sandboxing, keybindings, fast mode, output styles |
reference/authoring-skills.md | Build extensions: advanced skill frontmatter, dynamic context injection, skillOverrides, description-budget tuning, plugin-in-a-skill |
Highest-leverage, least-known (the entry points)
If unsure where to start, these are the features most likely to change how the user works:
/goal <condition> — keep running, re-checked after every turn, until a condition holds. The unattended-completion primitive. → autonomous.md- Channels permission relay — approve
Bash/Edit/Write from a phone (Telegram/iMessage). → remote-and-channels.md
- Routine
/fire endpoints — Claude Code as an HTTP webhook target for CI/alerting/deploys. → autonomous.md
claude --safe-mode — disable all customizations to debug a broken config without losing auth/model. → config-cli-env.md- Obscure hook events —
UserPromptExpansion, PostToolUse → updatedToolOutput, ConfigChange, InstructionsLoaded. → hooks.md
/btw — zero-cost side question with full context, no tools, never enters history. → context-and-session.md
