Skydive vs ZeroClaw

Skydive

Skydive publishes an OpenAPI 3.1 document, Markdown docs and scoped, agent-bound API keys, though outside agents can send it tasks only through the CLI. Its status page records several major incidents, including a 25-hour Agent Hosting outage, and its home and pricing pages say data is never used for training while the privacy policy turns training on by default.

ZeroClaw

Supervised autonomy is the default, so tools outside a short read-only list wait for the owner's approval, and secrets are encrypted in the config file. The docs state that command execution is not written to an audit log, and the project publishes no privacy policy or terms.

Skydive

1. Send tasks with skydive chat -p "..." --agent <id> --json and pass --conversation <id> from the JSON to keep context. The REST API can't send a task.
2. Don't retry chat -p or a create blindly after a timeout. There's no idempotency key, so list agents or keys before retrying.
3. Ask the owner for an agent-bound use key when you only need to read. Account-level edit keys can create agents and mint keys.
4. Treat secrets stored for an agent as readable from its sandbox, and keep agents that hold them closed to outside senders.
5. Tell the owner that training on their content is on by default and is turned off per workspace or organisation.

ZeroClaw

1. Keep [risk_profiles.<alias>] level = "supervised". full runs every tool not listed in always_ask, and the YOLO preset removes the sandbox and path rules too
2. Add sending and account tools such as send_via, google_workspace and http_request to always_ask so a session-level Always grant cannot skip the prompt
3. List senders in [peer_groups.<name>] external_peers. ["*"] accepts anyone on that channel
4. Set [web_fetch].allowed_domains and [http_request].allowed_domains. Fetches run unprompted and the OS sandbox does not block egress
5. Run one-shot tasks with zeroclaw agent -a <alias> -m "<task>". A cron or webhook turn has no prompt, so shell calls fall to the command policy