メニュー
Launch, monitor, stop, list, or resume long-running non-GPU local shell tasks without relying on one shell timeout. Use this when the user asks to run local build/release/maintenance commands that may outlive one shell call or one OpenCode session.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Operate the vh-solara agent harness itself — install, self-update, run the upgrade loop, read release migration notes, and use guide/update/doctor. Load this skill when asked to install, update, or operate the harness, or to read what changed in a release.
Backlog ledger discipline for vh-solara — conflict-safe edits to docs/planning/backlog.md (hybrid split-commit) plus DEFER/follow-up curation routing. Load this skill before editing the backlog, when handling a cas_conflict on the backlog, or when deciding where a DEFER/p2 finding should land.
Caller-facing routing aid for media-perception in vh-solara — decide between in-context perception (vision-capable caller) and single-delegation to the media-perception specialist (text-only caller or heavy/multi-step perception). Load this skill when you hold an image, diagram, chart, video, document/PDF, or audio locator and need to decide how to perceive it.
Debugging loop for existing bugs — build a deterministic red signal FIRST, then reproduce, minimise, hypothesise, instrument, fix, regression-test, with a 3-line post-mortem. Use this when fixing a bug, chasing a failure, or reproducing a defect. Includes an explicit downgrade-to-human handoff when deterministic red is unavailable.
Guide for creating and updating repo-local OpenCode skills in this coding repository. Use this whenever a user asks to create, update, improve, validate, or troubleshoot a skill under `.opencode/skills/`, or when archived sessions should be distilled into reusable coding workflows.
Test-driven development loop for new behavior — red, green, refactor, next vertical slice. Use this when adding or changing behavior test-first at pre-agreed seams. Loads when the user asks to TDD a feature, write tests first, drive an implementation with tests, or settle when refactor should happen.
| name | bgshell-job |
| description | Launch, monitor, stop, list, or resume long-running non-GPU local shell tasks without relying on one shell timeout. Use this when the user asks to run local build/release/maintenance commands that may outlive one shell call or one OpenCode session. |
| compatibility | opencode |
Use this for repo-local non-GPU shell work that may outlive one shell call or one OpenCode session.
/session-start <alias>), or pass --session <alias>.vh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py launch --job <name> -- <command ...>vh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py status --job <name> --lines 40vh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py logs --job <name> --lines 80vh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py stop --job <name> --forcevh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py listvh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py status --job-dir tmp/agent-runs/<alias>/bg-jobs/<name>vh-agent-harness exec python .opencode/skills/bgshell-job/scripts/bgshell_job.py resume --job-dir tmp/agent-runs/<alias>/bg-jobs/<name>State is saved under:
tmp/agent-runs/<session-alias>/bg-jobs/<job-name>/job.jsontmp/agent-runs/<session-alias>/bg-jobs/<job-name>/job.logqueued, starting, running, succeeded, failed, stopped, interrupted