Skip to content

Latest commit

 

History

History
27 lines (23 loc) · 2.73 KB

File metadata and controls

27 lines (23 loc) · 2.73 KB

Compatibility Matrix

Surface MVP behavior Notes
Codex $loop / $Loop Shipped first. $loop <objective> and the display-case $Loop <objective> map to the Codex skill in skills/loop/SKILL.md.
Loop Agent Console TUI Shipped keyboard-first baseline. loop with no arguments opens a terminal console with Tab/arrow focus navigation, Enter-opened run pickers and action menus, confirmation overlays for completion/Codex launch, two-field note input, wiki/log reading, Obsidian settings, lineage-preserving follow-up intent, dashboard open, and Codex resume actions when a session id is known.
Loop CLI Codex agent Prototype. loop run --agent codex "prompt" launches codex exec after Loop state and safety checks.
Loop CLI Claude Code agent Prototype. loop run --agent claudecode "prompt" launches claude --print after Loop state and safety checks.
Loop Wiki CLI Shipped local baseline. loop wiki list/read/open/serve reads .loop/wiki notes and serves a localhost dashboard.
Loop Wiki dashboard actions Shipped local baseline. Dashboard forms use persisted server-issued confirmation tokens for local mutations and external Codex terminal launch.
Loop Wiki Obsidian sync Shipped local filesystem baseline. loop wiki obsidian status/init/sync/watch/install-service mirrors only .loop/wiki/user/*.md into an explicit vault path under Loop/<project-name>-<project-id>/; derived AI/index/graph artifacts stay local and are regenerated after content imports. Delete and rename propagation is roadmap work.
Codex /goal Interop only. Long-running goal tracking can wrap the durable plan, but the core state remains local.
Codex automations Read-only or triage-only by default. Write-capable automation requires durable human approval.
Codex worktrees Supported as an isolation decision. Code-changing loops should prefer worktree or branch isolation.
Claude Code namespaced /loop Roadmap. A future native Claude command should default to a namespaced command to avoid conflicts.
Claude Code bare /loop Explicit opt-in only. Bare /loop can conflict with built-in or user-customized command behavior.
Connectors/MCP Roadmap. Optional after local durable state and first adapter tests are stable.
Issue tracker memory Roadmap. Local .loop state and Loop Wiki are the MVP baseline.

Command Naming

Codex uses $ for explicit skill invocation, so the first adapter targets $loop. Claude Code already has slash-command conventions, so the safe default for a future Claude adapter is namespaced rather than bare /loop.

The MVP documentation should never imply that Claude parity has shipped before the second adapter exists.