Security fixes are maintained on main. Feature and agent branches are temporary integration branches and are not release channels.

Do not open public issues containing credentials, tokens, private network details, customer information, exploit payloads, or operational secrets.

Report security issues privately to the repository owner through GitHub's private vulnerability reporting when enabled, or through an established private InnerChispa security contact.

Include:

• affected component and commit SHA;
• reproduction steps;
• impact;
• logs with secrets and personal data removed;
• suggested mitigation if known.

This repository is intended to contain reusable InnerOS platform code, public architecture documentation, tests, safe fixtures, and reproducible examples.

It must not contain:

• production credentials or tokens;
• private keys;
• customer data;
• personal phone numbers or personal email addresses used as runtime configuration;
• exact private LAN/Tailscale topology unless required for a sanitized example;
• live tunnel URLs tied to production;
• raw coordination data or agent inbox exports;
• production database dumps;
• deployment secrets.

Examples must use placeholders such as 192.0.2.10, example.com, and user@example.com.

All agent/tool execution must preserve:

• least privilege;
• explicit capability/tool authorization;
• server-derived identity;
• bounded filesystem and repository scope;
• immutable terminal task states;
• auditable evidence;
• human approval for destructive or high-impact actions;
• no secret material in model-visible logs.

Before merge to main, changes should pass:

• repository tests relevant to the touched component;
• secret scanning;
• dependency review where dependency manifests change;
• a diff review for accidental operational data disclosure.

Security-sensitive changes to OAuth, MCP authorization, task ownership, Temporal workflows, or sandboxing require an explicit verification note in the pull request.