Use to orchestrate a swarm of sub-agents under a CEO pattern when one agent isn't enough but a full visible grid is overkill: break a mission into roles, dispatch them, and coordinate via signal files.
Use to track and cap what an agent run costs: meter token/compute spend, set budgets, and surface cost as a first-class signal so an autonomous agent doesn't quietly burn through its limit.
Use to apply deterministic, zero-LLM code transforms: fast, repeatable edits that don't need a model, so an agent offloads mechanical changes to a cheap tool instead of spending tokens reasoning through them.
Use to author, test, and share agent skills from the command line: scaffold a SKILL.md, validate its structure, and package it for reuse, turning a one-off procedure into a portable capability.
Automatically loads AGENTS.md configuration file content at session start to ensure Claude Code follows project-specific agent behavior. Only loads if AGENTS.md exists, otherwise passes empty context. Supports the universal AGENTS.md standard for cross-platform AI assistant compatibility.
AI-powered bash command security guard. Before any Bash command runs, a lightweight Claude subagent evaluates it for destructive or irreversible patterns — recursive deletes, force pushes to protected branches, database drops, and credential exposure — and blocks execution with a clear explanation if flagged. Uses PreToolUse with type:agent, which is the only hook pattern that can block tool execution via AI reasoning.
Create automatic backup of files before any Edit operation for safety. This hook creates a timestamped backup copy (filename.backup.timestamp) of any existing file before Claude modifies it. Provides a safety net to recover previous versions if needed. Only backs up existing files, includes error suppression to handle edge cases gracefully.
Log every file mutation to CSV for demo prep. Records timestamp, tool, file path, action, and details for Edit, MultiEdit, Write, and Bash operations. Output: .claude/critical_log_changes.csv
Change Logger Hook Logs every file mutation (Edit, Write, Bash) to a CSV file for demo prep and session review. Output: .claude/critical_log_changes.csv
Real-time browser visualization of Claude Code's context window and subagent/tool execution as a git-graph timeline. Opens http://localhost:7878 on SessionStart with a vertical timeline (most recent on top), one column per agent (main + subagents branching from their Task tool call). Tool calls are color-coded by type (Read=green, Edit/Write=orange, Bash=red, Grep/Glob=cyan, Task=purple, Web=yellow, MCP=gray). Right sidebar shows context-window usage (tokens/200K) with cache_read/cache_creation/input/output breakdown plus per-subagent mini-context. Reads ~/.claude/projects/<encoded-cwd>/<session_id>.jsonl directly — no data replication, no network calls. Pure stdlib Python, zero pip dependencies. Persistent daemon HTTP server on port 7878 (auto-fallback 7879-7888 if busy; override with CONTEXT_TIMELINE_PORT env var). Watchdog auto-shutdown after 1h of inactivity. Disable browser auto-open with CONTEXT_TIMELINE_NO_BROWSER=1. Manual shutdown: python3 .claude/hooks/context-timeline.py --shutdown