@linktogo/maggie-workspace-bootstrap

Backs the maggie-workspace CLI from maggie: clones the configured repositories into a workspace, installs their dependencies cache-first, wires status hooks for a local coding agent — Claude Code or GitHub Copilot CLI — into each checkout, and maintains the shared board.json kanban state.

import { setSessionStatus } from '@linktogo/maggie-workspace-bootstrap';

await setSessionStatus('/ws/.maggie/board.json', 'example-api', 'sess-1', 'done');

Board states are todo, inprogress, question, and done. Each repo tracks one entry per agent session (keyed by the session id the hook payload carries); writes are atomic and keep a bounded per-session event history. Every session records the agent that produced it, set once.

agents.js is the registry the rest of the library reads: where each agent's settings file lives (.claude/settings.local.json, .github/copilot/settings.local.json), how to build its hook block, and how to read one back for drift detection.

Each session also tracks startedAt (set once) and usage — an { inputTokens, outputTokens, cacheCreationInputTokens, cacheReadInputTokens } object recomputed from the session's transcript on every end-of-turn event (Stop / agentStop), kept null until the first one. tokens.js holds both readers — Claude Code's transcript JSONL and Copilot's session-state/<id>/events.jsonl event stream, whose per-model inputTokens/outputTokens/cacheReadTokens/cacheWriteTokens are mapped onto the same four counters — plus the history.jsonl helpers.