At a glance
A proprietary coding harness from Random Labs for repository-scale terminal work, with provider and model-slot selection, granular tool permissions, session steering and resume, parallel and background subagents, reusable skills, MCP, long-running shells, headless structured output, and an HTTP server mode. Execution is host-first: permissions govern tool calls but the reviewed documentation does not establish an operating-system sandbox.
Good choice if
- Long repository tasks that need deliberate context management, project instructions, session continuation, and active steering
- Parallel research, implementation, and verification workflows using child sessions, background agents, and trace views
- A shared interactive and scripted workflow through the TUI, one-shot structured headless runs, queues, or the HTTP server
Check before choosing
- Slate is proprietary and the published npm package exposes no official public source repository, so HarnessMatch could not perform an immutable implementation audit
- Shells, workspace paths, and local MCP subprocesses execute on the host; no operating-system sandbox is documented, and --yolo bypasses every permission check
- Server mode binds to 0.0.0.0 by default according to the current documentation, so remote exposure requires deliberate password, network, and host controls
See 4 more considerations
- External workspaces, long-running shells, environment variables, skills, and MCP servers widen the trusted input and execution boundary and need separate review
- Documented tests, reviewer agents, trace cards, and changed-file summaries support inspection but do not independently prove task success or benchmark quality
- Session resume and queue persistence restore interaction state rather than rolling back files, shell effects, network actions, or external services; no checkpoint mechanism is documented
- The quickstart warns that Windows may not work and recommends WSL, and it reports significant terminal flicker when Slate runs inside tmux