菜单
Admin recovery for wedged Runs — preview the manager cursor, force a Station complete, set engine-managed fields, reset drift, or patch a feedback record. Every mutation requires explicit user confirmation.
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
Start a new darkrun Run — describe what you want to accomplish and the factory manager scaffolds a right-sized lifecycle for it
Create a composite Run combining stations from multiple factories with sync points
Configure darkrun for this project — auto-detect VCS, hosting, CI/CD, and default branch, confirm with the user, and write .darkrun/settings.yml
Parking lot for ideas not yet ready to become a Run — add, list, review, or promote a backlog item into a Run
Review and decide a Station's Checkpoint — approve to advance the Run, or request changes to route rework as drift
Run a Run lights-out in dark mode — pre-elaborate the work up front, then advance Station to Station without stopping for review, pausing only on external/await gates and ambiguity
| name | darkrun-debug |
| description | Admin recovery for wedged Runs — preview the manager cursor, force a Station complete, set engine-managed fields, reset drift, or patch a feedback record. Every mutation requires explicit user confirmation. |
Recovery for Runs the manager's normal loop can't clear. Call darkrun_debug with run + op (+ a reason on every mutating op), and do exactly what it returns. The tool confirms each mutation with the user before touching .darkrun/ — surface that prompt and never bypass or auto-retry it. Run the read-only preview_cursor op before and after any change.
When one Unit has run off the rails — burned its Pass budget (the manager escalates a runaway Pass loop), or its spec is wrong but its body is locked while it executes — reset just that Unit with darkrun_unit_reset (slug + unit, dry-run unless confirm:true). It returns the Unit to pending, which unlocks its body for editing and clears its Pass history, stamps, and gate results — preserving the spec (deps, inputs, outputs, declared gates, body). Then edit the spec and darkrun_tick: the Unit re-dispatches from Pass 1. This is lighter than a Station wipe (it touches one Unit, never its siblings) and is exactly what the runaway-Pass escalation message points you to.
The same capability is reachable from the desktop review UI without MCP: a unit-reset request there sets the Unit's reset_requested flag, and the engine performs the reset on its next tick.
For a destructive Station/Run wipe use /darkrun:darkrun-reset instead.
