autorun

autorun supplies task-lifecycle checks, command safety guards, skills, and native integration assets for AI coding harnesses. It is the only Python distribution here, and it contains the complete ar plugin plus the autorun and autorun-install entry points. pdf-extractor remains a separate harness plugin, but not a separate package: its extraction code ships inside this distribution, so install it with uv tool install 'autorun-ai[pdf]'.

Install

Install the published Python tool, then publish native assets for the harnesses detected on the machine:

uv tool install autorun-ai
autorun --install
autorun --status

Source checkout installation:

uv tool install 'git+https://github.com/ahundt/autorun.git#subdirectory=plugins/autorun'

Claude Code can instead install the plugin through its marketplace flow:

claude plugin marketplace add https://github.com/ahundt/autorun.git
claude plugin install ar@autorun

Use autorun --install --install-dry-run to preview filesystem and registration work. Use autorun --uninstall to remove only autorun-owned assets; session state, task history, and logs are retained intentionally.

Harness capabilities

The CLI accepts the shared command grammar internally and renders the spelling native to each harness: ar: for Codex, /ar- for ForgeCode and OpenCode, /ar for Pi, and /ar: for Claude and the Gemini-family harnesses.

Development

From the repository root:

uv sync --project plugins/autorun
uv run --project plugins/autorun pytest plugins/autorun/tests

Tests must redirect HOME, AUTORUN_HOME, AUTORUN_TEST_STATE_DIR, and AUTORUN_TEST_RUNTIME_DIR before importing autorun. Pi integration tests must also set PI_CODING_AGENT_DIR beneath the redirected HOME.

The project is licensed under the Apache License 2.0. See LICENSE in this distribution. Full project documentation is available at https://github.com/ahundt/autorun.