Shared memory, receipts, and an unlimited Free beta

Added

Shared Project Memory: Winskel keeps one memory per project and gives it to Claude Code and Codex before each run. Add facts yourself in Settings, Context and memory, approve what an agent proposes, and correct or forget any fact. Secrets are refused. Folder files can keep a copy in CLAUDE.local.md and AGENTS.md if you turn them on.
Execution receipts: every result shows what ran and on which Mac, the commands Winskel observed and who allowed each one, the checks Winskel ran or observed, and what was not verified. Chat answers say which agent and model answered.
Node.js and Git readiness: first run and Settings show whether Node.js and Git are ready on your Mac, with the fix and a Check again button.
Built, not verified: when Winskel cannot install, build or test the work on your Mac, the work is kept and labeled Built, not verified instead of Failed, and you can preview it.
Unlimited Free beta: no limit on projects, active models or routes during the beta.

Autonomous by default: there is no permission mode to choose. Routine work runs inside a sandbox without asking, and anything outside it, such as deleting files, pushing or reading secrets, still asks you first.
Streaming Chat: Claude Code answers appear as they are written, with a running timer.
Preview: Preview kept work straight from the conversation, and see which Mac holds the files when a result was made on another Mac.
Voice: dictated words appear in the message box as you speak.
Removing a project keeps your folder and its Git history, and says what is blocking the removal if something is.
Quieter results: the Codex note shows only when your request names Codex, and unused models are mentioned only when they explain a failure.

Removing a project could fail with "Could not remove the project" for projects that had run.
A missing Node.js was reported as "Claude Code was not available".
Work that could not be checked showed as a red Failed, with no way to preview it.
The agent's own Node.js and npm in standard install folders were blocked.
Some results showed contradictory status text, or two references for one run.

Claude Code and Codex run on your own accounts, and their usage limits still apply. When a plan reaches its limit, the task stops and you rerun it later; continuing on your other plan is on the roadmap.
Codex Chat answers arrive whole rather than streaming.
Work that is Built, not verified cannot be approved or applied until the missing tool is installed; preview it or choose Request changes.
Winskel runs on Apple Silicon Macs with macOS 12 or later.