Skip to content

Security: neyisyavuz-afk/local-agent-loop

Security

docs/security.md

Security and privacy

The repository is designed to be publishable without shipping live runtime state, credentials, or private environment details.

Repository contents

Included: orchestration source, tests, example configuration, helper scripts, and documentation.

Not included: model weights/runtime binaries, live sessions/ledgers/leases/logs, provider credentials/auth files, private hostnames, LAN/Tailscale addresses, or real user paths.

The example "apiKey": "local" value is a non-secret placeholder for a loopback local endpoint.

Trust boundaries

  • Model output is untrusted. Structured role output is validated before it can affect workflow authority.
  • Frozen task identity is authoritative. Task ID/revision/hash/scope/acceptance are control-plane owned.
  • Role contexts are isolated. Only bounded structured projections cross role boundaries.
  • GPU ownership is explicit. Local language and image workloads share one lease.

Deployment guidance

  • keep state outside product workspaces;
  • bind local runtimes to loopback unless you deliberately secure a network listener;
  • keep provider keys in environment/secret stores, never committed config;
  • review example paths before use;
  • never commit auth files, live ledgers, logs, screenshots, or private network details.

Pre-public scan

Run the read-only hygiene check:

pwsh ./scripts/scan-public.ps1

It checks tracked text files for common key/token patterns, personal emails/paths/private IPs, and Git history for non-noreply commit emails.

The scan is defense in depth, not a guarantee that arbitrary proprietary data is safe to publish. Review diffs before changing repository visibility.

There aren't any published security advisories