autonomous-research-loop
Run, maintain, or extend the paper-only autonomous research workflow for baseline, full experiment, and related comparison reporting across scripts, EXE packaging, startup flow, and operator panel act
SKILL.mdagents/skills/autonomous-research-loop/SKILL.md
GitHubAutonomous Research Loop
Binding sources
docs/FINAL_SYSTEM_VISION.md— continuous research / testnet data for L6 and L9–10; autonomy must stay bounded.AGENTS.md— paper-first; loops stop on operator/error/config; no unbounded hidden agents.
Purpose
Use this skill to manage the repository's autonomous paper-only research workflow.
This path supports research mode continuity and operator observability in docs/FINAL_SYSTEM_VISION.md (Layers 6, 9–10): long-running paper study, not live trading.
It covers the current research surface:
- baseline safe runs
- scorer experiment runs
- research-loop summaries and concentration reporting
- EXE build and Windows startup flow
- operator panel actions that trigger the same research paths
Use When
Use this skill when the task is one or more of:
- run or refine the autonomous paper-research loop
- improve baseline vs experiment orchestration
- improve compact experiment summaries
- package or launch the loop on Windows
- improve startup behavior, single-instance safety, or local research logs
- wire the same loop cleanly into the operator panel
Do Not Use When
Do not use this skill for:
- live trading automation
- exchange order execution changes
- hard risk weakening
- auto-tuning or auto-learning application
- strategy logic rewrites unrelated to research-loop operation
- operator-panel changes that do not touch research workflow behavior
Safety Contract
Never violate these rules:
PAPER_TRADING=trueremains the safe defaultAI_MODE=OFF,LEARNING_MODE=OFF, andSCORER_CONFIDENCE_EXPERIMENT=falseremain repo defaults- baseline and experiment flows stay clearly separated
- experiments stay opt-in and labeled
- no hidden runtime behavior change
- no secret logging or secret packaging
Primary Workflow
- Read
AGENTS.md. - Inspect the current research scripts, reporting path, EXE helpers, and any touched operator-panel action.
- Preserve the baseline path first.
- Make the smallest safe orchestration or reporting change.
- Run the quality gate if scripts or code changed.
- Verify baseline path still works.
- Verify experiment or trial path still works if touched.
- Summarize the operational outcome, not a file dump.
Required Checks
- Baseline and experiment outputs remain distinguishable.
- Trial labels remain explicit in logs, summaries, and panel views.
- EXE or startup helpers do not mutate repo defaults.
- Research summaries stay compact and decision-oriented.
- Windows helpers remain readable, local-safe, and paper-only.
Expected Output
What Changed:file-level summary.Autonomous Workflow:what runs in what order now.Trial Visibility:how baseline and experiment stay distinct.Safety Check:explicit confirmation that defaults and hard controls remain intact.