AGENT SKILL

migrate-to-workbox

Offload a running local herdr session to the workbox - relocate an idle agent's in-flight work to the remote box through a handoff so it continues on fast hardware. Use when the user wants to migrate, move, or offload local herdr agents/sessions to the workbox or remote box, dogfood the migration, or get work off the slow-network laptop.

License 许可
MIT
所属仓库
Necmttn/skills
最近更新
2026-11-07
Stars
★ 0
Forks
0

01 前言 Front Matter

name
migrate-to-workbox
description
Offload a running local herdr session to the workbox - relocate an idle agent's in-flight work to the remote box through a handoff so it continues on fast hardware. Use when the user wants to migrate, move, or offload local herdr agents/sessions to the workbox or remote box, dogfood the migration, or get work off the slow-network laptop.

03 SKILL.md 正文

Migrate a local herdr session to the workbox

Offload: take an agent running locally and relocate its work to the workbox - the persistent remote herdr server on fast hardware - so the laptop's slow link stops gating it. The carrier is a handoff: a committed document plus a GitHub issue, machine-independent so any successor can resume from it.

This skill is the workbox-specific playbook. For the socket primitives (agent list/read/send, pane send-keys, agent start) and the meaning of drive and handoff, see herdr-agent-orchestration - don't restate them here.

Preconditions

  • The workbox is reachable over ssh and runs a herdr server (ssh <workbox> 'herdr status' shows running).
  • The agent CLI you'll start there is installed and authed on the workbox; gh works there.
  • You can drive both servers over the socket - local directly, workbox via ssh <workbox> 'herdr …'.

Steps

  1. Pick targets. herdr agent list locally. Migrate only idle agents - a working agent has no stable state to hand off. List the chosen panes before touching any.
  2. Source writes the handoff (per agent, one at a time). Drive the local agent to: push its branch, write a handoff document at the repo root, open a GitHub issue whose body is that document, and reply with the issue URL. Capture {repo, branch, checkpoint commit, issue URL}. Done when: the issue URL is posted and the branch is pushed.
  3. Prepare the workbox checkout. On the workbox, ensure the repo is present (clone it if missing) and fetch the handoff branch. If another agent already occupies that repo's working tree, give this one its own git worktree - never two agents in one tree.
  4. Start the successor on the workbox. ssh <workbox> 'herdr agent start <name> --cwd <repo-or-worktree> -- <agent-cli> <bypass-flag>'. Clear the startup trust/hook prompts once it boots. Give each successor its own labeled workspace (--workspace <id>, or pane move --new-workspace --label <repo> after) - never pack agents into one workspace, it makes sessions unreviewable. Done when: agent list on the workbox shows the new pane idle.
  5. Hand it the goal. Send the successor the handoff issue as a tracked goal - /goal <objective referencing the issue and branch> - not a one-shot prompt. Done when: a read shows it pursuing the goal.
  6. Retire the local source. Tell the local agent its work has moved to the workbox (or close its pane) so the same work never runs in two places.
  7. Repeat per target. The workbox server persists every successor; the human watches them with herdr --remote <workbox> --session <name> and detaches without stopping them.

With Syncthing (live file sync)

If the workbox keeps the project folder live-identical to the laptop (e.g. via Syncthing, as in backpine/remote-agent-workspace), step 3 simplifies: the repo and the branch's files are already on the box, so skip the clone/fetch. The git branch + issue handoff still earns its place - it carries the tracked goal and survives a fresh context and per-machine reauth, neither of which file sync provides. Two agents in the same synced repo still need separate git worktrees.

Gotchas

  • One handoff at a time: let each source finish posting its issue before starting its successor, so you never mix up {repo, branch, issue}.
  • Two targets sharing one repo (e.g. two agents in the same project) must land in separate worktrees on the workbox.
  • The successor reauths per machine and reads the issue, not the laptop's memory - so it resumes cleanly even with a fresh context.

04 同仓库更多 Skills

Necmttn
Necmttn/skills

wiki

Recall from the personal LLM wiki at ~/wiki. Use when the user asks what the wiki says, wants prior thinking/intuition on a concept, or when grounding a design or decision in previously synthesized knowledge. Read-only - never writes to the wiki.

★ 0
Necmttn
Necmttn/skills

wiki-ingest

Curate a source into the personal LLM wiki at ~/wiki following its SCHEMA.md - a file, URL, research output, ax session, or a pending ~/wiki/inbox/ brief. Use when the user says "add this to the wiki", "ingest this", "promote the inbox", or after research worth keeping long-term.

★ 0
Necmttn
Necmttn/skills

wiki-harvest

Propose-only sweep of recent coding sessions (via the ax CLI) for recurring concepts that are missing from, or should update, the personal LLM wiki at ~/wiki. Writes candidate briefs to ~/wiki/inbox/ only - never wiki pages. Use on schedule or when the user asks "harvest the wiki" or "what concepts am I repeating".

★ 0
Necmttn
Necmttn/skills

no-ai-slop

Strip AI-slop patterns from a draft while preserving the writer's voice, or detect slop without rewriting. Use as the mandatory final pass on any copy before shipping, when the user wants a draft clearer/less AI-sounding, or when asked "is this AI slop?", "audit this writing", "does this read as AI".

★ 0
Necmttn
Necmttn/skills

aso-appstore-screenshots

Generate high-converting App Store screenshots by analyzing your app's codebase, discovering core benefits, and creating ASO-optimized screenshot images using Nano Banana Pro.

★ 0
Necmttn
Necmttn/skills

app-idea

Run a new or existing consumer/AI app idea through the golden-track playbook (~/wiki/playbook/apps.md) from intake to verdict, write a verdict card, stop for the owner, and ratchet owner overrides back into the doctrine. Use when the user shares an app idea, asks "should I build X", "audit <app>", "re-score <app>", or pastes an idea in a product context. Never builds; never promotes a rule on its own.

★ 0