Agent skill · luongnv89
opencode-handoff
Resume a limit-blocked OpenCode session in a fresh sandbox: same project and agent setup, no host OpenCode config/token/key, attached in a tmux panel. Don't use for one-shot sandbox tasks (opencode-sandbox), opencode.ai (opencode-runner), or Herdr.
What it needs
About 5k tokens when loaded.
What this skill does
OpenCode Handoff You hit the OpenCode usage limit mid-task. This skill gives you a new local OpenCode profile for the same project with the same agent setup, deliberately leaves the host's OpenCode config, token, and key behind, and opens it in a new tmux panel. The credential exclusion is the point, not a side effect. Anonymous/free models may work without an account, and authentication is optional when the selected provider requires it. A fresh local profile does not guarantee a fresh provider, model, account, network, or service allowance. This skill composes two others rather than reimplementing them: Concern Owned by --- --- Container lifecycle, mounts, credentials opencode-sandbox (runopencode.sh) Panel mechanics tmux-agent-comms (Phase 1 — create/discover) Credential boundary + wiring the two together this skill When to Use OpenCode reports a usage or rate limit and you want to keep going now You want a second OpenCode session on the same project under a different account, without disturbing the host session You want a disposable OpenCode environment that cannot spend the host account Do not use it to run a one-shot task in a container (that is opencode-sandbox), to call opencode.ai's hosted models (opencode-runner), or to manage Herdr panes (herdr-agent-comms). Repo Sync Before Edits (mandatory) The container edits the project tree through the /workspace mount, and it only sees what is on disk at docker run time. Sync before launching: If the tree is dirty: stash (-u), sync, pop. If origin is missing or the rebase conflicts, stop and ask the user — never rebase blind on a dirty tree. Because the handoff container writes to the same working tree as the host session, decide with the user which session owns the tree before starting work in both. What crosses into the container, and what does not This is the whole design. Read references/credential-boundary.md before changing any mount. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills luongnv89/opencode-handoff