### Run updated package setup with autoswarm --home Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/setup.md Run the updated package's setup command to refresh a host installation. Reuse custom --skill-dir, --settings-file, and --plugin-dir values. Invoking setup through the old installed launcher does not obtain the new package. ```bash autoswarm --home setup --host ``` -------------------------------- ### Install AutoSwarm with git clone and setup for Codex Source: https://github.com/hkuds/autoswarm/blob/main/README.md First installation steps for the current host using git clone and uv. Reuse an existing checkout when available. Other host IDs include claude-code, opencode, cursor, copilot, muse, codebuddy, hermes, nanobot, dsh, grok. Without uv, use a dedicated Python 3.10+ environment and python -m pip install . ```bash git clone https://github.com/HKUDS/AutoSwarm.git cd AutoSwarm uv tool install --force . autoswarm setup --host codex ``` -------------------------------- ### Update existing AutoSwarm installation with git pull and reinstall Source: https://github.com/hkuds/autoswarm/blob/main/README.md Update an existing Git checkout, preserving local edits and collecting active delegated work. Run on the intended branch. Use the actual host, existing AutoSwarm home and original installation destinations. Put --home before setup when needed; retain custom --skill-dir, --settings-file and --plugin-dir values. ```bash git pull --ff-only uv tool install --reinstall . autoswarm setup --host codex ``` -------------------------------- ### Run updated package setup with python -m autoswarm Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/setup.md Alternative invocation of the updated package's setup command using the module form. Pass the same arguments as the autoswarm command, including --home and --host. ```bash python -m autoswarm ``` -------------------------------- ### Install or Update AutoSwarm for an Agent Source: https://github.com/hkuds/autoswarm/blob/main/README.md Prompt to give to an existing agent to install or update AutoSwarm, reuse existing configuration, and verify locally installed agents. The assistant will report which collaborators are ready and can handle adding new agents by name. ```bash Install or update AutoSwarm for this agent. Reuse my existing configuration and verify which locally installed agents are available. ``` -------------------------------- ### Install pre-push hook Source: https://github.com/hkuds/autoswarm/blob/main/CONTRIBUTING.md Installs a repository-local pre-push hook that runs the checks before pushing. The hook uses the Python interpreter that installed it; commit or stash changes first so it validates the exact commit being pushed. ```bash python scripts/check.py --install-hook ``` -------------------------------- ### Start a Task with AutoSwarm Source: https://github.com/hkuds/autoswarm/blob/main/README.md Prompt to give to an agent to start a task, such as reading documents and summarizing key approaches and disagreements. AutoSwarm selects, connects, and coordinates the right agents for the task. ```bash Use AutoSwarm to read these 100 documents and summarize the key approaches and disagreements. ``` -------------------------------- ### Reinstall with uv tool install --reinstall . Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/setup.md Use this command to reinstall AutoSwarm from an updated checkout when the original installation method was uv. It forces a reinstall even if the development version label has not changed. ```bash uv tool install --reinstall . ``` -------------------------------- ### Set up development environment with uv Source: https://github.com/hkuds/autoswarm/blob/main/CONTRIBUTING.md Creates a Python 3.12 virtual environment, installs the package in editable mode, and verifies the CLI. Requires Python 3.10+; on Windows executables are under Scripts/. ```bash uv venv .venv-v2 --python 3.12 uv pip install --python .venv-v2/bin/python -e . .venv-v2/bin/autoswarm --version .venv-v2/bin/autoswarm --help ``` -------------------------------- ### Collect download statistics with TRAFFIC_TOKEN Source: https://github.com/hkuds/autoswarm/blob/main/docs/download-statistics.md Run the collector script with the repository, start date, and output directory. Requires TRAFFIC_TOKEN in the process environment. Keep the same output directory's history.json for subsequent runs; for the original baseline, copy docs/download-statistics-baseline.json there as history.json before the first collection. Repository and start date must match existing history. Writes local files only; the workflow publishes them. ```bash python scripts/collect_downloads.py --repository HKUDS/AutoSwarm --since 2026-10-10 --output /path/to/statistics ``` -------------------------------- ### Reinstall with python -m pip install --force-reinstall . Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/setup.md Use this command to reinstall AutoSwarm from an updated checkout when the original installation method was pip, using the existing dedicated Python environment. It forces a reinstall even if the development version label has not changed. ```bash python -m pip install --force-reinstall . ``` -------------------------------- ### Run complete CI checks Source: https://github.com/hkuds/autoswarm/blob/main/CONTRIBUTING.md Runs the full validation suite: builds and installs the wheel in a temporary venv, checks entry points, runs Python and Node tests, and verifies a relocatable Skill outside the checkout. Use a Python 3.10+ interpreter (preferably 3.12) and Node.js 20. ```bash python scripts/check.py ``` -------------------------------- ### Parent delegation commands with communication rules Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/communication.md Commands for the parent to delegate work to child agents, setting contacts and communication rules. Names are scoped to the assigning parent and may refer to peers started afterward. Retain the returned call_id, stable agent_id, and controller_id. ```text run --agent codex --name implementation --allow-peer review --communication-rules "Share interface changes with review; own the implementation." --input implementation.txt --background run --agent cursor --name review --allow-peer implementation --communication-rules "Ask implementation about interfaces; own the review." --input review.txt --background ``` -------------------------------- ### Verify providers with autoswarm doctor Source: https://github.com/hkuds/autoswarm/blob/main/README.md Verifies requested providers in the actual project, replacing the example provider and directory as appropriate. Repeat --agent for several providers. This spends model calls to check a reply and native resume; verified: true confirms the probe. Version checks only establish executable discovery. ```bash autoswarm doctor --agent codex --probe --cwd /path/to/project ``` -------------------------------- ### Run Python and Node test suites Source: https://github.com/hkuds/autoswarm/blob/main/CONTRIBUTING.md Runs the Python unit tests and the OpenCode plugin tests. These require local socket and subprocess access; Node is needed only for the OpenCode plugin tests. ```bash python -m unittest discover -s tests -v ``` ```bash node --test tests/test_opencode_plugin.mjs ``` -------------------------------- ### Run and resume with model and effort overrides Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/models.md Use with the AutoSwarm CLI to select a model and reasoning effort for a run or resume. Omission preserves native defaults; resume inherits previously explicit choices unless supplied values replace them. Internal message turns and idle-message continuation retain the selections. ```bash autoswarm run --agent codex --model --effort --input task.txt --background ``` ```bash autoswarm resume --effort --input followup.txt ``` -------------------------------- ### Answer a native question with answer.json Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/interactions.md After reading native_request, response_format, response_help, user_required, and delegate_to_parent, write the answer JSON using the supplied format and run this command. The original request stays with its child connection and native turn. ```text answer --input answer.json ``` -------------------------------- ### Child communication commands Source: https://github.com/hkuds/autoswarm/blob/main/src/autoswarm/assets/references/communication.md Commands for children to send, reply, wait, receive, inspect, and deliver messages. Targets are scoped names, stable agent IDs, returned call IDs, or 'parent'. A reply uses the received ID and returns to its sender without requiring a reverse peer grant. ```text send review --input question.txt send implementation --interrupt --input urgent-question.txt reply msg_ID --input answer.txt wait --message msg_ID receive inspect msg_ID deliver msg_ID ``` -------------------------------- ### Cite AutoSwarm Paper Source: https://github.com/hkuds/autoswarm/blob/main/README.md BibTeX entry for citing the AutoSwarm paper in research. Includes title, authors, year, and eprint identifier. ```bibtex @misc{li2026autoswarm, title={AutoSwarm: Automating Agent Swarms}, author={Zhonghang Li and Zongwei Li and Chao Huang}, year={2026}, eprint={2610.33048} } ```