Fullhouse  issues → tested draft PRs from inside Claude, ChatGPT or Grok  ·  add it →

Threads

A thread is one chat with one agent: its own model, mode (plan / auto / full), and native agent session underneath.

Start one

  • New chat: Cmd+N, or pick a project in the left sidebar.
  • The composer carries agent, model, and mode — type / for commands, @ to mention files.
  • Esc stops a running turn. Shift+Tab toggles plan mode. ↑/↓ recalls composer history.

Exact resume

Each thread tracks its agent's exact native session ID (--resume, exec resume, -s depending on the agent). Reopening a thread resumes that session — Fullhouse never guesses "latest session".

Threads persist as one JSON file per chat under the OS app-data path: inspectable, greppable, backed-up. Not browser localStorage.

Move work to another agent

When a provider rate-limits, a banner offers Continue in <agent>: the transcript is extracted, secrets redacted (see Permission modes), the git worktree preserved, and the fallback agent starts from the same state. The target thread inherits the source worktree, and the switch is recorded in a signed handoff ledger.

Next: Permission modes — what agents may do.