@verkflode/remit-claude-code
v0.4.1
Published
Remit for Claude Code: Allow/Ask/Never authority over every tool call via a PreToolUse hook, with job grants and a shared decision log.
Readme
@verkflode/remit-claude-code
Remit for Claude Code: personal agent authority — owner-set Allow / Ask / Never tiers enforced on the PreToolUse hook seam, with every decision recorded.
The tier mapping is exact: Allow → allow (no prompt), Ask → ask (Claude Code's own native approval prompt is the transport), Never → deny with a reason the model sees.
Install
npm install -g @verkflode/remit-claude-code
remit-cc initinit creates ~/.remit/ with conservative defaults and prints the hook snippet to add to your Claude Code settings (.claude/settings.json per project, or ~/.claude/settings.json for everywhere). Remit deliberately never edits settings files itself — the hook is the authority surface, so installing it is the owner's move.
Owner surface
remit-cc status tiers and state paths
remit-cc grant <areas> [scope:<path>] [for:<dur>] pre-authorize a bounded batch
remit-cc grants / revoke <id|all> list / end grants
remit-cc cap <area> <N> per <hour|day|7d> temporal cap (v0.4)
remit-cc cap raise <area> to <N> for <dur> temporary, self-expiring raise
remit-cc caps caps and current usage
remit-cc export --dogwood-events decision log as a Dogwood-replayable traceState lives in ~/.remit/ — the machine-wide authority root shared by every Remit adapter, so a grant or cap set here binds every governed agent on the machine, and the decision log answers "what did ANY agent do today" in one place.
Seam grade and honest limitations are stated in design 0002: hooks are the outermost interception this host offers (the catalog classifies settings/hook files as change-settings, default Never, so governed tools cannot remove the hook), and ask resolutions are not reported back to hooks.
MIT © Verkflöde
