2026-04-29 TIPLLM Robot Learning Handoff

Summary

Rene requested that TIP crawler/robot planning use TIPLLM only, no external AI layer, and that crawler/robot experience be written into a Gitea training pool so TIPLLM improves over time.

Implemented locally and synced to Erik:

Verification robot controller with safe profiles:
- erik-safe
- pi-fetch
- proxmox-heavy
TIPLLM planning mode:
- --tipllm-plan --limit=N
Robot experience writer:
- status snapshots
- TIPLLM plans
- queue dry-runs
- queue enqueues
- rejected queue plans
- future crawler results
Learning pool import:
- TIP_TRAINING_REPO/qa-pairs/**/*.jsonl is imported into the tip_llm lane by learning-pool:build.

Safety Choices

Default robot profile is erik-safe.
erik-safe is capped to 3 lightweight queues by default.
Playwright and discovery queues are excluded from erik-safe.
No crawler jobs were started during this work.
No queue waves were enqueued during this work.

Training Pool

The local Gitea training pool clone already existed at /tmp/tip-training-data.

A first robot experience was written and pushed:

f1c83f8 crawl: add robot-status training records [2026-04-29T20:11:24.091Z]

Erik

Synced code/docs to /opt/tip and ran the scraper TypeScript build.

The build initially failed because two stale misplaced remote-only duplicate files existed on Erik:

/opt/tip/packages/scraper/src/scrapers/scheduler.ts
/opt/tip/packages/scraper/src/vendor-discovery-crawler.ts

They were not present locally and had wrong relative imports. Removed only those duplicates. Build passed after that.

PM2 services were left running and were not restarted.

Commands To Start Safely

npm run robots:verification -w packages/scraper -- --status
npm run robots:verification -w packages/scraper -- --tipllm-plan --limit=3
npm run robots:verification -w packages/scraper -- --enqueue=details-fast-lane --profile=erik-safe --dry-run

2.0 KiB Raw Blame History