Skip to content

Other-computer handoff

This is the portable starting point for Shadow on another Mac.

Bootstrap

bash
git clone --branch shadow-v1.0.1 --depth 1 \
  https://github.com/firstbitelabsllc/shadow.git
cd shadow
bash install.sh
shadow doctor

install.sh links the command and mounts the same skill in each native host root that exists (~/.claude/skills, ~/.agents/skills, ~/.cursor/skills). Pass --no-skills to link the command alone.

Expected result: shadow doctor reports product identity, command, and the host/mount checks that exist on that computer. Missing optional hosts or mounts are warnings; Cursor cold directive activation is an explicit unsupported surface. Authentication stays inside the native host on that computer.

The normal loop

  1. Establish one stable seat name and run shadow status --by <seat>. The computer board owns global project priority, entity pointers, claims, owners, leases, and entity resume checkpoints.
  2. Resume that seat's claims, then read the selected entity's committed PLAN.md, which owns its milestones, checkpoints, detail, and proof.
  3. Atomically claim and dispatch path-disjoint checkpoints through supported native hosts. Freeze each handoff and allow only the exact paths it may change.
  4. Review the diff and reproduce the important test locally. A host receipt is evidence, not acceptance by itself.
  5. Record each result, uncertainty, proof, blocked wake, and reachable successor in the owning PLAN.md; continue until full acceptance or hard rails.

Do not create another queue, autonomous router, daemon, watcher, cloud executor, credential relay, transcript store, or parallel status database. A foreground, explicit role choice may be used locally, but it never launches or substitutes a host. Keep evidence inside the project-local .shadow/evidence/ path and never put credentials, prompts, raw transcripts, provider payloads, or absolute private paths in it.

Main skill map

SurfaceUse it for
/shadowStart/resume work, read the Outcome, drain reachable lanes, and leave proof plus successors.

shadow amp is a CLI subcommand, not a mounted skill: it projects a paste-ready resume block for a checkpoint the seat already claimed — see amp. Two optional extension slots fill routed recall (memory) and finish-quality-plus-voice (taste) when mounted — see extension slots. The pre-mortem slot is gone: its pre-commit timing has no shipped mechanism (a repo may declare an adversarial-lens slug — post-work review, a different moment — or a person may delegate foresight to their own method). Diagrams stay Brief-contract law, no slot involved.

Use native Codex, Claude Code, or Cursor for execution. Provider-specific helpers are adapters; none becomes the plan authority or stores credentials.

Read the public state

  • Repository: firstbitelabsllc/shadow
  • Do not trust a version or commit copied into this guide. Read the protected main branch and release tags before starting work.
  • Use git ls-remote origin refs/heads/main refs/tags/shadow-v*, then read VERSION from the exact checkout or release tag you chose.
  • Run shadow doctor on that checkout. The result describes that computer's mounts and host tools; it is not a remote-host claim.

Fast local readback

bash
git fetch origin main --tags
git merge --ff-only origin/main
git status --short --branch
shadow doctor

When handing work to the next computer, pass the repository, exact revision, entity plan pointer, claimed checkpoints, allowed paths, proof commands, and blocked wakes. That is enough context; do not paste a transcript or copy the other computer's root board.