rendo-aclip

rendo-aclip is the canonical Python SDK for ACLIP, the Agent Command Line Interface Protocol.

It keeps normal CLI usage natural while standardizing the parts agents actually depend on:

• progressive Markdown help
• app-defined success output plus structured error envelopes
• sidecar manifests for distribution metadata
• packaging helpers for shipping runnable CLI artifacts

Install

Canonical package:

pip install rendo-aclip

Short-name official alias:

pip install aclip

Both install paths are first-party and synchronized. The import path is the same either way:

from aclip import AclipApp

If you want the canonical dependency name in project manifests, prefer rendo-aclip. If you want the shortest install command, aclip is the official alias.

Smallest End-to-End CLI

main.py

from aclip import AclipApp


def create_app() -> AclipApp:
    app = AclipApp(
        name="notes",
        summary="A minimal notes CLI.",
        description="Create and list notes from a small local CLI.",
    )

    def create_note(title: str, body: str) -> dict:
        """Create a note in a local JSON store.

        Args:
            title: Title for the note.
            body: Body text for the note.
        """
        return {"note": {"title": title, "body": body}}

    app.group(
        "note",
        summary="Manage notes",
        description="Create and inspect notes.",
    ).command(
        "create",
        handler=create_note,
        examples=["notes note create --title hello --body world"],
    )

    return app


app = create_app()

cli.py

from aclip import run_cli
from main import app


run_cli(app)

If you prefer lazy initialization at process start, the launcher also accepts the factory directly:

from aclip import run_cli
from main import create_app


run_cli(create_app)

Run it like a normal CLI:

python cli.py --help
python cli.py note --help
python cli.py note create --help
python cli.py note create --title hello --body world

The final command prints app-defined success output:

• strings stay plain text
• objects and arrays are emitted as plain JSON

AclipApp.name is the canonical CLI command token, not a display title. Keep it executable-safe, with no spaces. version is optional during local authoring. Set it before manifest build, packaging, publish, skill export, or the default root --version / -V / -v surface, which prints <cli-name> <version>. --help / -h remain protocol-reserved discovery flags, while --version / -V / -v are only default root fallbacks and become author-owned again after a concrete command path is selected. When one author-owned argument should accept multiple aliases, declare flags=(\"--long\", \"-S\") on ArgumentSpec instead of splitting aliases across multiple parameters.

Build A Distributable CLI

From a dedicated build script:

import aclip


artifact = aclip.build("main:app")

print(artifact.binary_path)
print(artifact.manifest_path)

build(...) is the shortest first-class path. "main:app" is the runtime import target the packaged binary will execute. That is why the recommended pattern is a separate build.py script instead of having the app object “build itself”.

If you prefer to keep initialization behind a function, ACLIP also supports an explicit factory target:

import aclip


artifact = aclip.build(factory="main:create_app")

Python also supports a shorthand when you already imported a top-level factory:

import aclip
from main import create_app


artifact = aclip.build(create_app)

If you want the fully explicit name, build_cli(...) is the same API.

Authentication

ACLIP now standardizes a minimum auth contract around portable credential declarations and an optional reserved auth control plane.

from aclip import (
    AuthCommandConfig,
    AuthNextAction,
    AuthStatus,
    CredentialSpec,
    auth_status_result,
    build_auth_control_plane,
)


credentials = [
    CredentialSpec.env(
        name="notes_token",
        env_var="ACLIP_NOTES_TOKEN",
        description="Remote notes API token.",
        required=True,
    ),
    CredentialSpec.file(
        name="notes_token_file",
        path=".secrets/notes-token.txt",
        description="Optional local token file.",
    ),
]

auth = build_auth_control_plane(
    AuthCommandConfig(
        login_description="Login to the author-defined remote service.",
        login_examples=["notes auth login"],
        login_handler=lambda _payload: {"status": "logged_in"},
        status_description="Inspect current auth state.",
        status_examples=["notes auth status"],
        status_handler=lambda _payload: auth_status_result(
            AuthStatus(
                state="authenticated",
                principal="dev@rendo.cn",
                next_actions=[
                    AuthNextAction(
                        summary="Refresh before expiry",
                        command="notes auth login",
                    )
                ],
            ),
            guidance_md="Credential is valid. Refresh before running long-lived jobs.",
        ),
        logout_description="Logout from the author-defined remote service.",
        logout_examples=["notes auth logout"],
        logout_handler=lambda _payload: {"status": "logged_out"},
    )
)

Diagnostics

ACLIP also provides a reserved optional doctor control plane and helper payload builders for stable agent-facing checks.

from aclip import (
    DoctorCheck,
    DoctorCommandConfig,
    DoctorRemediation,
    build_doctor_control_plane,
    doctor_result,
)

doctor = build_doctor_control_plane(
    DoctorCommandConfig(
        check_description="Run author-defined environment checks.",
        check_examples=["notes doctor check"],
        check_handler=lambda _payload: doctor_result(
            checks=[
                DoctorCheck(
                    id="credentials",
                    status="warn",
                    severity="medium",
                    category="auth",
                    summary="Credential is missing or expired.",
                    hint="Run the login flow before retrying.",
                    remediation=[
                        DoctorRemediation(
                            summary="Refresh the credential.",
                            command="notes auth login",
                            automatable=True,
                        )
                    ],
                )
            ],
            guidance_md="Fix the auth check first, then rerun the original command.",
        ),
        fix_description="Apply author-defined fixes for failed checks.",
        fix_examples=["notes doctor fix"],
        fix_handler=lambda _payload: {"checks": []},
    )
)

Export Agent Skills Packages

ACLIP can export developer-authored skill packages while keeping CLI and command metadata aligned as anchors.

There are two different hook scopes:

• add_cli_skill(...) Attach one skill package to the whole CLI. Use this for top-level usage strategy, cross-command navigation, auth/doctor preparation, or other whole-tool guidance.
• add_command_skill(...) Attach one skill package to one command path. Use this for command-specific best practices, prerequisites, recovery steps, or multi-step invocation guidance.

These hooks do not change normal runtime behavior.

They do not automatically print skill text when the CLI runs. They only register export-time linkage so that export_skills(...) knows which developer-authored packages to materialize.

from pathlib import Path

from aclip import export_skills

from main import create_app


app = create_app()
skills_root = Path("skills")
app.add_cli_skill(skills_root / "notes-overview")
app.add_command_skill(
    ("note", "create"),
    skills_root / "note-create-best-practice",
    metadata={"owner": "docs"},
)

artifact = export_skills(app, output_dir=Path("dist") / "skills")
print(artifact.index_path)

What happens here:

1. app.add_cli_skill(...) registers one whole-CLI package
2. app.add_command_skill(...) registers one command-bound package
3. export_skills(...) validates each source package
4. ACLIP copies those packages into the output directory
5. ACLIP injects anchor metadata into SKILL.md
6. ACLIP writes a skills.aclip.json index beside the exported packages

Each source package must contain a developer-authored SKILL.md.

Important boundary:

• hook registration does not install anything into a user workspace
• export_skills(...) is an explicit optional API
• if you want init or another command to copy exported skills into a project, that copy step is still your command logic, not automatic ACLIP runtime behavior

In a conventional project layout, ACLIP infers:

• project root
• source root
• executable name

src/ is optional. Advanced overrides such as project_root, source_root, and extra_paths are still available for monorepos or non-standard layouts, but they are no longer the default path.

What You Get

• AclipApp for tree-shaped CLI authoring
• direct handler=... registration and decorator authoring
• app.run(...) for direct execution in tests or custom hosts
• run_cli(...) for the default launcher path without manual sys.argv[1:]
• build(...) for the shortest packaging path, with build_cli(...) as the explicit equivalent
• export_skills(...) for Agent Skills-compatible package export from CLI-level and command-level hooks

When To Use ACLIP

Use rendo-aclip when you want a CLI that still feels natural to command-line users while giving agents:

• predictable help disclosure
• predictable machine-readable failures
• a stable packaging and distribution path

If your goal is only a human-first CLI with free-form text output, ACLIP is probably more structure than you need.

Python vs TypeScript

The Python and TypeScript SDKs now share the same primary story:

• define app in main
• launch with run_cli(app) / runCli(app)
• package with build("main:app")

One intentional difference remains: Python also supports build(create_app) for a top-level factory function. TypeScript does not expose the same shortcut because a JavaScript function object is not a stable packaging import target on its own.

Repository

• https://github.com/rendo-studio/aclip