Server Observation Loop
Workflow
- Ask for current goal and available context.
- Produce one minimal command or script for Wyatt to run.
- Tell Wyatt exactly what output, screenshot, or metric to paste back.
- Analyze returned evidence: observed facts, likely causes, next actions.
- Update RUN_CARD, DATA_CARD, ERROR_CARD.
- Never request raw confidential data.
Hard rules
- Give one minimal command or script at a time, with exact output/screenshot/log evidence to paste back.
- Never request raw confidential/company data, credentials, cookies, tokens, or full private datasets.
- Base analysis only on returned evidence; separate observed facts from inference.
- Update RUN_CARD, DATA_CARD, or ERROR_CARD for every non-trivial training/debug loop.
Output style
- What I see
- What I infer
- What you should run next
- What to paste back