ccc — Call Coding CLIs

A Rust library and CLI for calling coding agents (opencode, claude, codex, kimi, roocode, crush) from your Rust programs or terminal. Run prompts against any supported coding agent with a consistent interface, streaming support, and structured output parsing.

Quick Start

Install the CLI

cargo install ccc

Rust Library

[dependencies]
ccc = "0.1"

use ccc::{CommandSpec, Runner};

let runner = Runner::new();

// Build and run a command
let spec = CommandSpec::new(["opencode", "run", "Explain this code"])
    .with_cwd("/path/to/project");

let result = runner.run(spec);
println!("{}", result.stdout);

CLI Usage

# Run a prompt with opencode (default)
ccc "fix the bug in main.rs"

# Use a different runner
ccc "review my changes"           # opencode (default)
ccc cc "review my changes"         # claude
ccc cx "write tests"               # codex
ccc k "explain this function"      # kimi

# Thinking level (+0 to +4)
ccc +3 "design the API"           # highest thinking effort

# Model selection
ccc :claude-4 "hello"              # provider:model syntax
ccc : Sonnet 4 "hello"            # just model name

# Preset alias from config
ccc @reviewer "current changes"

# Force literal prompt (no flag parsing)
ccc -- "use --yolo in the prompt"

# Streaming output
ccc ..fmt "build a CLI"

Library API

CommandSpec

Describes a command to run: argv, stdin, working directory, and environment.

use ccc::CommandSpec;

// Minimal
let spec = CommandSpec::new(["opencode", "run", "hello"]);

// With options
let spec = CommandSpec::new(["claude", "-p", "review this"])
    .with_cwd("/home/user/project")
    .with_stdin("context here")
    .with_env("ANTHROPIC_API_KEY", "sk-...");

CompletedRun

Result of a completed command run.

let result = runner.run(spec);
println!("stdout: {}", result.stdout);
println!("stderr: {}", result.stderr);
println!("exit code: {}", result.exit_code);

Runner

Executes commands with blocking or streaming modes.

use ccc::Runner;

let runner = Runner::new();

// Blocking run
let result = runner.run(spec);

// Streaming with callback (stdout/stderr per line)
let result = runner.stream(spec, |channel, chunk| {
    if channel == "stdout" {
        print!("{}", chunk);
    } else {
        eprint!("{}", chunk);
    }
});

build_prompt_spec

Convenience helper to build an opencode run command from a prompt string.

use ccc::build_prompt_spec;

let spec = build_prompt_spec("fix the off-by-one error")?;
let result = Runner::new().run(spec);

CLI Reference

Runner Selectors

Thinking Levels

+0 / +none   Disable thinking
+1 / +low    Low effort
+2 / +med    Medium effort (default for most)
+3 / +high   High effort
+4 / +max    Maximum effort

Model Selection

ccc :provider:model "prompt"     # Full provider:model
ccc :model "prompt"              # Model only (uses default provider)

Output Modes

Shortcuts: .text, ..text, .json, ..json, .fmt, ..fmt

Permission Modes

--permission-mode safe   # Ask before dangerous ops (default)
--permission-mode auto  # Auto-approve safe ops
--permission-mode yolo  # Skip all permission checks
--permission-mode plan  # Plan-only mode
-y / --yolo             # Shortcut for --permission-mode yolo

Other Flags

--show-thinking         Show thinking in formatted output
--no-show-thinking      Hide thinking in formatted output
--sanitize-osc          Strip OSC escape sequences (titles, bells)
--no-sanitize-osc       Keep OSC escape sequences
--forward-unknown-json  Forward unparseable JSON lines to stderr
--print-config          Print example config file
--                       Treat all following args as literal prompt

Configuration

Config file lookup order:

1. .ccc.toml in project directory (or nearest ancestor)
2. XDG_CONFIG_HOME/ccc/config.toml
3. ~/.config/ccc/config.toml
4. $CCC_CONFIG (explicit path)

Generate an example config:

ccc --print-config

Example .ccc.toml:

[defaults]
runner = "oc"
provider = "anthropic"
model = "claude-4"
thinking = 2
show_thinking = false
sanitize_osc = true
output_mode = "formatted"

[abbreviations]
mycc = "cc"

[aliases.reviewer]
runner = "cc"
provider = "anthropic"
model = "claude-4"
thinking = 3
show_thinking = true
output_mode = "formatted"
agent = "reviewer"
prompt = "Review the current changes"
prompt_mode = "default"

Aliases

Aliases bundle runner, model, thinking level, and prompt into a single name:

ccc @reviewer "current changes"

The @reviewer alias might resolve to claude with specific settings and a base prompt like "Review the current changes". Your explicit prompt is appended or prepended based on prompt_mode.

Prompt Modes

• default — Use user prompt if provided, else alias prompt
• prepend — Add user prompt before alias prompt
• append — Add user prompt after alias prompt

Environment Variables

Supported Runners

License

Unlicense