Claude Built in commands and skills

#claude #claudecode #generative #ai

1- Core Built-In Commands (Engine & Session Controls)

Built-in commands execute instantly without sending a message to the model. They give you direct control over memory, model choices, and terminal execution.

Context & Memory Management

/context : Opens a visual map showing context window usage. Use it to see what percentage of token budget is consumed by project files vs. conversation history.
/compact [instructions] : Summarizes earlier turns in the chat to reclaim token space. You can append instructions like /compact preserve database migration notes.
/clear : Wipes the current session history clean for a new task while keeping global memory files (CLAUDE.md) intact.
/memory : Directly views and edits active CLAUDE.md guidelines.

Session Flow & Navigation

/plan [description] : Switches Claude into Plan Mode. Prevents file modifications while Claude drafts a step-by-step strategy for high-level tasks.
/btw [question] : Asks a side question without adding it to the main conversation context or history.
/rewind (alias /checkpoint) : Rolls back file changes, conversation turns, or both to a prior checkpoint.
/branch [name] / /fork : Forks the current conversation into a new branch so you can experiment with alternative implementations without destroying your main timeline.

Execution Controls

/model [model] : Hot-swaps models mid-session (e.g., jump from Sonnet for fast refactoring to Opus for complex math/architecture).
/effort [level] : Adjusts reasoning depth (e.g., low, medium, max) depending on task complexity.
/cost / /usage / /stats : Displays API token expenditures, daily rate-limit statuses, or long-term trends.

2- Bundled & Custom Skills (AI Workflows)

Skills are intelligent workflows. When you invoke a skill, it loads a structured prompt with specific tools, allowing Claude to execute multi-step logic or spawn parallel subagents.

Essential Shipped Skills

/simplify [focus] : Spawns multiple review agents to analyze recently changed code for redundancies, edge-case bugs, or performance gains, then applies cleaner refactors automatically.
/code-review / /security-review : Runs automated multi-agent checks on pending diffs to identify security risks, anti-patterns, or subtle bugs.
/batch [description] : Decomposes large-scale, repetitive changes across an entire repository (e.g., "migrate 40 endpoint signatures to use Zod") into independent units running in separate worktrees.
/loop [interval] [prompt] : Periodically runs a task on a schedule (e.g., /loop 2m "check if the local build passes").
/debug [description] : Reads system runtime/error logs to run structured troubleshooting.

3. How Skills Are Configured Under the Hood

When building custom skills in .claude/skills//SKILL.md or .claude/commands/, you control execution behavior via YAML Frontmatter:

---
name: Database Migration Helper
description: "Generates and validates Prisma/Knex migration files"
# Restrict tools to prevent destructive shell commands
allowed-tools: ["FileEdit", "Bash(npm run migrate:*)"]
# Isolate execution to a subagent to avoid cluttering main session context
context: fork
agent: Plan
# Control who triggers the skill
disable-model-invocation: true # Can ONLY be triggered manually by user via /
user-invocable: true # Visible in / autocomplete menu
---
Analyze the latest database schema updates in `src/db/schema.ts`.
Create a migration file under `src/db/migrations/` and test it using `npm run migrate:test`.

Frontmatter Directives Key:

context: fork: Runs the skill inside an isolated background subagent so your main chat remains clean.
disable-model-invocation: true: Prevents Claude from running this skill automatically during regular text turns—it will only run when you explicitly type /migration.
allowed-tools: Scopes down security permissions so the skill can only execute approved commands or file actions.