yaacli 0.76.0

TUI reference implementation for ya-agent-sdk

YAACLI CLI

TUI reference implementation for ya-agent-sdk.

Usage

Run with uvx:

uvx yaacli

Install with uv:

uv tool install yaacli
yaacli

Update with uv:

uv tool upgrade yaacli

Install with pip:

pip install yaacli
yaacli

Run as a module:

python -m yaacli

Built-in Skills

YAACLI ships with building-agents from the repository canonical source skills/agent-builder/.

The YA Claw deployment skill lives in skills/ya-claw-deploy/ and is published as YA_CLAW_DEPLOY_SKILL.zip during release.

The repository sync script keeps bundled skill files under packages/yaacli/yaacli/skills/ aligned.

Development

This package lives in the ya-mono workspace.

git clone [e-mail]:YOUR_NAME/ya-mono.git
cd ya-mono
uv sync --all-packages
cp packages/yaacli/.env.example packages/yaacli/.env

YAACLI loads .env from packages/yaacli/.env and the current working directory. Provider API keys can live in that .env file or in ~/.yaacli/config.toml under [env]. SDK and tool variables such as YA_AGENT_* and search API keys can also live in that same .env file because YAACLI loads it into the process environment at startup. Use packages/ya-agent-sdk/.env.example as the reference list for SDK and tool variables.

Codex OAuth credentials can be created once and reused from YAACLI:

uvx ya-oauth login codex

Then set model = "oauth@codex:gpt-5.5" in a YAACLI model profile.

Model profiles are configured in ~/.yaacli/config.toml and selected with /model inside the TUI:

[general]
model = "anthropic:claude-sonnet-4-5"
model_settings = "anthropic_adaptive_high"
model_cfg = "claude_200k"

[model_profiles.fast]
label = "Fast"
model = "openai-responses:gpt-5-mini"
model_settings = "openai_responses_low"
model_cfg = "gpt5_270k"

[model_profiles.codex_oauth]
label = "Codex OAuth"
model = "oauth@codex:gpt-5.5"
model_settings = "openai_responses_high"
model_cfg = "gpt5_270k"

[general] is the startup fallback profile. The last selected profile is remembered in ~/.yaacli/state.json and restored on the next launch when that profile still exists.

Shell command review is configured in ~/.yaacli/config.toml under security.shell_review:

[security.shell_review]
enabled = true
model = "gateway@openai-responses:gpt-5.4-mini"
model_settings = "openai_responses_low"
on_needs_approval = "defer"
risk_threshold = "high"

When enabled, model is required. model_settings accepts SDK preset names or an inline TOML table. risk_threshold defaults to high and controls when the configured action triggers.

Run CLI tests from the workspace root:

make test-cli

Clipboard Image Paste

Plain terminal paste always inserts text into the input box. Use Ctrl+V or /paste-image to attach an image from the system clipboard. On macOS terminal apps over SSH, map Command+Shift+V to send Ctrl+V if you want a native-feeling shortcut.

YAACLI reads clipboard images through Pillow first on macOS and Windows. macOS also reads Finder-copied image files through Cocoa pasteboard APIs via pyobjc-framework-Cocoa. Linux image paste still relies on wl-paste on Wayland or xclip on X11.

License

BSD 3-Clause License. See the repository license.