This repo is a template for running powerful local agents. That means small mistakes can become dangerous if the template encourages bad defaults.

Please report issues related to:

• secrets handling
• accidental credential exposure
• unsafe default permissions
• prompt or tool-policy bypass paths
• watchdog / scheduler behavior that can create duplicate consumers or unintended background execution
• channel integrations that can leak messages or attachments

The template is designed around a few principles:

• Secrets live outside the repo whenever possible
• Runtime logs are gitignored
• Messaging credentials are stored under ~/.claude/channels/..., not in the project
• The agent should run as a single long-lived session, not a swarm of duplicate consumers
• Hooks, watchdogs, and scheduling should be inspectable and repairable

Please do not open a public issue for a credential leak or security bug that could affect real users.

Instead, report it privately to the maintainer with:

• what component is affected
• exact reproduction steps
• impact
• suggested mitigation if you have one

If you suspect compromise:

1. Rotate any bot/API tokens immediately
2. Stop the running agent session
3. Disable launchd jobs or other schedulers
4. Review runtime logs under .claude/runtime/
5. Remove any accidentally committed secrets from history before publishing

These are usually not security vulnerabilities by themselves:

• local-only debugging logs that are already gitignored
• ws:// or cleartext traffic on a private LAN unless it crosses a trust boundary
• behavior that only occurs after the operator explicitly disables safeguards