# AGENTS

- Repo state: currently docs/templates/notebook only (`AGENTS.md`, `poc_eval_sheet_guide.md`, `poc_eval_sheet_template.csv`, `poc_eval_notebook.ipynb`); there are still no manifests, source code packages, tests, or CI configs to infer tooling from.
- Project intent: ML-assisted surveying for reconstruction/demolition photos, identifying objects and their material composition.
- Current phase: Google Colab proof of concept focused on validating the ML pipeline (not production app architecture).
- Expected pipeline concept: surveyor taps object -> segmentation isolates target -> model predicts object + material composition -> structured material inventory for circular construction.
- Collaboration workflow: start by confirming the real goal and core decision to drive, then work in small buckets; for each bucket, provide a plan and pause for checkpoint review before continuing.
- Decision hygiene: explicitly restate/verify key decisions during execution to avoid drift from the agreed goal.
- Update this file once code/config lands; prefer executable repo sources over prose when they conflict.