菜单
Start or resume execution of a Nightshift shift by forking into the manager subagent. Use when the user invokes /nightshift:start.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Add a task definition to an existing Nightshift shift, registering it in manager.md and adding a status column to table.csv. Use when the user invokes /nightshift:add-task.
Archive a Nightshift shift by moving its directory to .nightshift/archive/YYYY-MM-DD-<name>/. Use when the user invokes /nightshift:archive.
Create a new Nightshift shift — a structured unit of batch agent work — with manager.md and an empty table.csv. Use when the user invokes /nightshift:create.
Execute one Nightshift task on one item, self-validate, retry on failure, and emit a structured result. Invoked by manager via `claude -p` subprocess; runs in a fresh top-level Claude Code session and inherits all user-configured MCPs.
Verify that Nightshift's required system dependencies (qsv, flock, jq) are installed and on PATH. Use when the user invokes /nightshift:doctor.
Run a single Nightshift task on a single table item for testing — without modifying table.csv or manager.md. Spawns a claude -p subprocess of /nightshift:do-task with --read-only. Use when the user invokes /nightshift:test-task.
| name | nightshift:start |
| description | Start or resume execution of a Nightshift shift by forking into the manager subagent. Use when the user invokes /nightshift:start. |
| disable-model-invocation | true |
| context | fork |
| agent | manager |
| allowed-tools | Bash(qsv *) Bash(flock *) Bash(test *) |
| argument-hint | ["shift-name"] |
Execute Nightshift shift $ARGUMENTS.
Run the bundled pre-flight script to check shift state and emit a JSON summary:
${CLAUDE_SKILL_DIR}/scripts/preflight.sh $ARGUMENTS
If the pre-flight reports any of these conditions, stop without forking the manager and report the message back to the user:
error: shift_not_found → "Shift $ARGUMENTS does not exist. Use /nightshift:create $ARGUMENTS first."error: no_items → "Shift $ARGUMENTS has no items. Use /nightshift:update-table $ARGUMENTS to add items."error: no_tasks → "Shift $ARGUMENTS has no tasks. Use /nightshift:add-task $ARGUMENTS to add tasks."status: complete → "Shift $ARGUMENTS is already complete. Use /nightshift:archive $ARGUMENTS to archive it."${CLAUDE_SKILL_DIR}/scripts/preflight.sh $ARGUMENTS --human
Read the manager.md and table.csv inside .nightshift/$ARGUMENTS/. Process all remaining items autonomously following the orchestration logic in your system prompt.
.nightshift/$ARGUMENTS/.nightshift/$ARGUMENTS/manager.md (read for task order, parallel config, disable-self-improvement flag).nightshift/$ARGUMENTS/table.csv (read for item statuses; do not write status transitions — the dev subagent does that).nightshift/$ARGUMENTS/.envRun all todo items to completion (or failure after retries), apply step improvements as you go, and emit the final shift-complete summary when done.