메뉴
Live flowai-workflow run supervisor. Use only for requests to run, monitor, continue, or resume a current workflow now. Exclude last-night, finished, post-mortem, root-cause-report, and conceptual questions.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Supervisor for one flowai-workflow run (Codex). Starts or resumes a single workflow, diagnoses failures from run artifacts, patches root causes, resumes the same run. Run inside an isolated Codex worker subagent spawned by the `supervise`/`orchestrate` skill.
Copy a bundled flowai-workflow into the user's project so they can adapt and run it locally. Use when the user wants to start using flowai-workflow in a new repo.
Run the flowai-workflow orchestration policy loop. Delegates policy execution to the workflow orchestrator agent, which selects workflows from ORCHESTRATION.md and supervisor summaries.
Execute a bundled or project-local flowai-workflow DAG. Use to launch a workflow run from inside the host IDE.
Long-cycle flowai-workflow orchestrator (Codex). Reads project orchestration policy, selects the next workflow, and returns a structured supervisor delegation request. Run inside an isolated Codex worker subagent spawned by the `orchestrate` skill.
Scaffold or adapt flowai-workflow DAGs in a project. Use for adding flowai-workflow, adapting an existing .flowai-workflow/<name>, validating workflow.yaml, or wiring agent prompts/scripts.
SOC 직업 분류 기준
| name | supervise |
| description | Live flowai-workflow run supervisor. Use only for requests to run, monitor, continue, or resume a current workflow now. Exclude last-night, finished, post-mortem, root-cause-report, and conceptual questions. |
| argument-hint | workflow folder path (e.g., .flowai-workflow/github-inbox) |
| effort | high |
Drive one flowai-workflow run to a clear terminal state by delegating the run
to supervisor.
Launch supervisor with the host's real subagent mechanism:
Agent / Task tool with
subagent_type=supervisor;@supervisor <task prompt> mention
syntax.worker subagent (the parent dispatches; Codex
max_depth=1 forbids nested spawns) and instruct it — by skill name — to
invoke the supervisor skill and return the SUPERVISOR_REPORT block as
its final message. The worker, not the parent, reads all run artifacts.Pass:
--resume <run-id>, if the user provided one;Ask the supervisor to return only a concise stop report: workflow, run id, final state, failed/stalled node, evidence paths read, fix surface, resume command, and remaining blocker.
If the host has neither a native subagent mechanism (Claude subagent_type,
OpenCode @mention) nor Codex-style worker dispatch, say that
context-isolated supervision is unavailable in this IDE and stop. Do not
silently run the full diagnosis loop in the parent context. Do not simulate
delegation with bash, echo, heredocs, or placeholder text.
The parent session may locate the requested workflow folder, but must not:
.flowai-workflow/ORCHESTRATION.md;flowai-workflow directly unless the user explicitly rejects subagent
delegation and asks for inline supervision.orchestrate.scaffold.