gh-timurkhakhalev-cc-plugins-headless-cli-agents/references/agent-specific-commands.md

CLI Agent Commands Reference

Claude Code CLI (Anthropic)

Basic Headless Usage

claude -p "Your prompt here"

Key Flags

• -p or --prompt: Execute one-shot prompt and exit
• --add-dir <path>: Add additional directory to workspace context
• --model <model>: Specify model (optional)

Examples

# Basic one-shot query
claude -p "Explain this function"

# Pipe input to Claude
cat error.log | claude -p "Summarize these errors"

# Multiple directory context
claude -p "Review the API design" --add-dir ../api-specs

Authentication

Requires configured API key or OAuth token. Run claude --help for setup options.

---

OpenAI Codex CLI

Basic Headless Usage

codex exec "Your prompt here"
codex e "Your prompt here"  # Short alias

Key Flags

• --full-auto: Unattended operation with workspace-write sandbox
• --dangerously-bypass-approvals-and-sandbox or --yolo: Complete hands-off mode (use carefully)
• --skip-git-repo-check: Allow execution outside Git repositories
• --cd <path>: Set working directory
• --model <model> or -m: Specify model (e.g., -m gpt-5-codex)

Examples

# Automated refactoring
codex exec --full-auto "Update all README links to HTTPS"

# Outside Git repo
codex exec --skip-git-repo-check --full-auto "Create hello world HTML"

# Different working directory
codex exec --cd /path/to/project "Fix failing tests"

Input Methods

# Pipe prompt from file
codex exec - < prompt.txt

# Standard input
echo "Review this code" | codex exec -

---

Google Gemini CLI

Basic Headless Usage

gemini --prompt "Your prompt here"
gemini -p "Your prompt here"  # Short form

Key Flags

• --prompt or -p: Execute prompt and exit
• --output-format <format>: Output format (json, stream-json)
• --model <model>: Specify model variant

Examples

# Basic query
gemini -p "Summarize API design in this repo"

# Pipe input with prompt
echo "List TODO comments" | gemini -p "-"

# JSON output
gemini -p "Analyze code structure" --output-format json

# Process file with instruction
cat DESIGN.md | gemini -p "Improve this design document"

Authentication

Requires Google account authentication or API key setup.

---

OpenCode CLI

Basic Headless Usage

opencode -p "Your prompt here"
opencode --prompt "Your prompt here"

Key Flags

• -p or --prompt: Execute single prompt and exit
• -f <format> or --format: Output format (json)
• -q or --quiet: Suppress loading spinner
• --cwd <path>: Set working directory

Examples

# Basic query
opencode -p "Explain Go context usage"

# JSON output
opencode -p "How many files in project?" -f json

# Quiet mode for scripting
opencode -p "Review code" -q

# Different working directory
opencode -p "Analyze this project" --cwd /path/to/project

Environment Setup

Requires API keys for providers (OpenAI, Anthropic, etc.) in environment variables.

---

Alibaba Qwen Code CLI

Basic Headless Usage

qwen -p "Your prompt here"

Key Flags

• -p or --prompt: Execute one-shot prompt
• --output-format <format>: Output format (json)
• --model <model>: Specify Qwen model variant
• --yolo: Bypass confirmations (similar to other agents)

Examples

# Code review
qwen -p "Review this code for potential bugs"

# Generate tests
qwen -p "Generate unit tests for utils.py"

# Pipe diff for review
git diff | qwen -p "Review this diff for errors"

# JSON output
qwen -p "List project files" --output-format json

Authentication

• First-time setup: Run qwen interactively to login with Qwen.ai OAuth
• Cached credentials: Used automatically for subsequent -p calls
• Local models: Set OPENAI_API_KEY and related env vars for local LLM servers

---

Factory Droid CLI

Basic Headless Usage

droid exec "Your prompt here"

Key Flags

• --auto <level>: Set autonomy level (low, medium, high)
• --skip-permissions-unsafe: Bypass all permission checks (use carefully)
• --cwd <path>: Set working directory
• -f <file>: Read prompt from file
• -o <format>: Output format (json)

Autonomy Levels

• Default (no flag): Read-only mode, safe for analysis
• --auto low: Allow low-risk file edits (documentation, simple refactors)
• --auto medium: Allow development operations (install packages, run tests)
• --auto high: Permit production-level changes (full access)

Examples

# Read-only analysis
droid exec "List all TODO comments across the project"

# Low-risk edits
droid exec "Fix typos in README.md" --auto low

# Development operations
droid exec "Fix failing unit tests" --auto medium

# High-risk changes
droid exec "Implement OAuth2 migration" --auto high

# Read prompt from file
droid exec -f prompt.md

# JSON output
droid exec "Analyze codebase" -o json

# Different working directory
droid exec "Review this code" --cwd /path/to/project

Safety Notes

• Default mode is read-only for safety
• Use --skip-permissions-unsafe only in sandboxed environments
• Consider autonomy levels carefully based on use case

---

Common Patterns

piping Input

Most agents support piping input:

# Pipe file content
cat file.txt | agent -p "Process this"

# Pipe command output
git log --oneline | agent -p "Summarize commits"

Reading from Files

# Direct file reading (if supported)
agent -f prompt.txt

# Using cat and pipe
cat prompt.txt | agent -p "-"

JSON Output

For scripting and automation:

# JSON output format
agent -p "Query" --output-format json
agent -p "Query" -f json
agent -p "Query" -o json

Automation Flags

For completely unattended operation:

# Various automation flags per agent
codex --full-auto "Task"
droid --auto medium "Task"
gemini --yolo "Task"  # if available