From dolphin-flow-harness
Dolphin Flow Harness Autopilot - state/pipeline strict execution mode
How this skill is triggered — by the user, by Claude, or both
Slash command
/dolphin-flow-harness:dfh-autopilotThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
[DFH AUTOPILOT ACTIVATED]
[DFH AUTOPILOT ACTIVATED]
Execute DFH work strictly through the persisted autopilot state and pipeline. Do not run a free-form workflow. Do not skip stages. Do not invent transitions.
state file: .dfh/autopilot/state.jsonpipeline paths from state:paths.specFilepaths.planFilepaths.openQuestionsFilecurrent stage: pipeline.stages[pipeline.currentStageIndex]required completion signal: currentStage.signalIf these conflict with chat history, state wins.
active.currentStage.signal.Modification Points before any completion signal.Modification Points must include - (none yet): <reason>.Stages and canonical signals:
analyst (phase expansion) -> DFH_STAGE_ANALYST_COMPLETEplan (phase planning) -> DFH_STAGE_PLAN_COMPLETEexecute (phase execution) -> DFH_STAGE_EXECUTE_COMPLETEqa (phase qa) -> DFH_STAGE_QA_COMPLETEverify (phase validation) -> DFH_STAGE_VERIFY_COMPLETEqa and verify may be skipped by pipeline config. If skipped in state, do not run them.
specFile.openQuestionsFile.specFile.planFile.planFile.When working, structure output as:
## DFH Autopilot - <stage id>
- State file: <path>
- Current phase: <phase>
- Required signal: <signal>
- Progress: <what was completed>
- Modification Points:
- <file or module>: <what changed>
- <file or module>: <what changed>
- Remaining: <what is still pending>
- Blockers: <none | details>
Modification Points is mandatory in every stage report.
If there are no file edits yet, use:
- Modification Points:
- (none yet): <reason>
When stage is complete, append the exact signal in its own line:
DFH_STAGE_<...>_COMPLETE
Only after all non-skipped stages are complete:
complete.Do not claim completion early.
npx claudepluginhub dolphin57/dolphin-claude-code --plugin dolphin-flow-harnessCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.