claude-code

name	claude-code
description	Reference for using Claude Code CLI to handle complex coding tasks. Use when Research-Claw needs to delegate multi-file projects, code review, refactoring, or agentic coding workflows to Anthropic's Claude Code. Trigger words: claude code, claude cli, delegate coding, code agent.

Claude Code CLI (Anthropic)

Agentic coding tool by Anthropic. Reads, edits, and runs code with full project context. Supports non-interactive print mode for programmatic use.

Detect Installation

which claude  # should return a path
claude --version

Install

npm install -g @anthropic-ai/claude-code

Requires authentication: run claude auth login (Anthropic account or --console for API billing).

Non-Interactive Usage (print mode)

For programmatic invocation from Research-Claw, use -p (print mode):

# Basic non-interactive execution
claude -p "Refactor the auth module to use async/await"

# Pipe content for analysis
cat src/utils.ts | claude -p "Find bugs in this file"

# JSON output for parsing
claude -p --output-format json "List all TODO comments in the project"

# Skip permission prompts (for fully automated pipelines)
claude -p --dangerously-skip-permissions "Fix all lint errors"

# Budget and turn limits
claude -p --max-turns 5 --max-budget-usd 2.00 "Add unit tests for the auth module"

# Model selection
claude -p --model sonnet "Explain this codebase architecture"

Key Flags

Flag	Description
-p, --print	Non-interactive mode: run task, print result, exit
--model <model>	Set model (sonnet, opus, or full model ID)
--output-format <fmt>	text / json / stream-json
--max-turns <n>	Limit agentic turns (print mode)
--max-budget-usd <n>	Spending cap in USD (print mode)
--dangerously-skip-permissions	Skip all permission prompts
--allowedTools <tools>	Auto-approve specific tools, e.g. "Bash(git *)" "Read"
--tools <tools>	Restrict available tools, e.g. "Bash,Read,Edit"
--add-dir <path>	Grant access to additional directories
--append-system-prompt <text>	Append to default system prompt
-c, --continue	Continue most recent conversation
-r, --resume <id>	Resume session by ID or name
--effort <level>	low / medium / high / max

Research-Claw exec Patterns

# Delegate a complex coding task
exec("claude -p --dangerously-skip-permissions --max-turns 10 'Implement the feature described in TODO.md'")

# Read-only analysis with budget cap
exec("claude -p --tools 'Read' --max-budget-usd 1.00 'Review code quality and list issues'")

# JSON output for structured results
exec("claude -p --output-format json 'List all exported functions in src/'")

Docs

• CLI reference: https://code.claude.com/docs/en/cli-reference
• Common workflows: https://code.claude.com/docs/en/common-workflows