Edge Pipeline Orchestrator coordinates the edge research stages that the sibling skills handle one by one. Starting from a tickets directory or an OHLCV file, it runs candidate detection, hint extraction, concept synthesis and strategy drafting, then reviews the drafts and exports the eligible ones.
The review loop runs for at most two iterations. Drafts that still need revision after that are downgraded to research probes. A manifest records every stage. It handles research files and gives no investment advice.
What You Get
- A pipeline_run_manifest.json with the full execution trace
- Folders for tickets, hints, concepts, drafts and exportable tickets
- Review output for each iteration of the review loop
- strategies/<candidate_id>/ folders holding strategy.yaml and metadata.json for exported drafts
Common Use Cases
- Running every edge research stage end to end on one day’s data
- Resuming from existing drafts to review, revise and export them
- Doing a dry run to preview which drafts would be exported
What Makes It Unique
It is the only skill in the edge group that chains all the others and keeps a bounded review and revision loop with fixed rules for downgrades and export eligibility. In Claude Code it can also take hints that Claude wrote itself through a file option. It needs the sibling skills checked out together, because it finds them by repository path.
Under The Hood
The agent runs one Python script that starts the five sibling skills’ scripts as subprocesses, reads local ticket, OHLCV and draft files, and writes YAML, JSON and strategy folders under the output directory. No API keys are needed and the skill description names no web calls. Needs Python and PyYAML, and the sibling skill folders must be present. Reference files on the pipeline flow and revision rules load when needed.