The supported source is main at version 0.1.0. There is no older release
line and no packaged binary channel.
This file is a reporting path, not a claim that Meshloop is production-hardened. The threat model records design requirements; it is not a guarantee that the current scaffold enforces them.
- It does not store credentials. It uses CLI sessions you already authenticated. Do not put API keys in config, issues, or reports.
- It does not install a Windows service or run as a daemon.
- It does not merge onto the current branch unless you pass
integrate --accept-integrate. - Live workers require
--allow-live-harness. Completing a live Claude Code, Codex, Pi, Grok, or Agy dispatch is not an R1 stamp.
Private reports are appropriate for:
- Path escape from a worker worktree or
allowed_paths - Secrets or personal data written into
.meshloop/state.sqlite, logs, or fixtures - Unsolicited live harness spawn without
--allow-live-harness - Worktree or git operations that touch the operator branch without an explicit integrate gate
- Redaction failures in
--jsondiagnostics
- “The fixture harness is not Claude / Codex”
- Missing WSL2, macOS, Linux, packaging, or concurrency greater than 1
- Model quality or plan quality
- Documented R1 limitations (see ADR 0016)
Use GitHub private vulnerability reporting.
Do not open a public issue or pull request with exploit detail, credentials, session transcripts, or private prompts. Redact secrets from any logs you attach.
There is no published security email and no bug bounty. The maintainer is @smota. There is no response SLA.