OpenClaw instance requirements

What to add to the OpenClaw instance

OpenClaw should orchestrate the early product workflow and PulseLine experience.

Workspace project directoryCreate a dedicated workspace repo/folder for centering code, test images, rules profiles, evaluation scripts, and generated reports.

Credentials and config entriesAdd protected config for artifact storage, optional API keys, model endpoints, queue settings, and TCGMarketPulse integration URL. Do not store grader content or user images in skills.

Image-processing runtimeInstall Python dependencies such as OpenCV, Pillow, NumPy, scikit-image, and optional ONNX Runtime/PyTorch for segmentation. Keep this in an isolated venv or container.

Artifact directoriesCreate durable folders/buckets for originals, normalized crops, overlays, debug masks, and JSON reports with retention policy.

Background worker pathUse a queue/worker for analysis so Discord/PulseLine does not block while images process. For prototype, a local command wrapper is enough.

PulseLine command surfaceAdd a conversational pattern: user sends photo → skill validates → worker analyzes → PulseLine replies with short result + overlay.

Evaluation harnessAdd repeatable tests against a golden image set. Each model/rules change should produce a before/after accuracy report.

Operational dependencies

Prototype-ready checklist.

Python 3 environment
opencv-python-headless
Pillow
NumPy
scikit-image
FastAPI or equivalent local service
SQLite/PostgreSQL connection for analysis records
Local or object-storage artifact path
Optional ONNX/PyTorch segmentation runtime

OpenClaw data flow

How a PulseLine analysis should move through the instance.

Discord / TCGMarketPulse upload
  → OpenClaw receives image attachment
  → pokemon-card-centering-intake skill validates request and creates analysis job
  → centering-vision-worker runs quality checks, card detection, rectification, measurement
  → centering-grader-rules skill compares measurements to PSA/CGC/BGS/SJC profiles
  → centering-report-writer skill creates short chat response + detailed report + overlay
  → PulseLine replies with result and stores artifacts for later review

Config needed

Names are illustrative; exact OpenClaw config shape can follow the running instance conventions.

Config item	Purpose	Example value
CENTERING_WORKDIR	Where worker reads/writes local artifacts	/root/.openclaw/workspace/tcg-centering
CENTERING_ARTIFACT_RETENTION_DAYS	Retention window for user-uploaded images	30
CENTERING_RULES_PATH	Versioned grader profile directory	rules/grader-profiles
CENTERING_WORKER_TIMEOUT_SECONDS	Analysis timeout before user gets retry message	45
TCGMARKETPULSE_API_URL	Optional product backend integration	https://…/api
CENTERING_MODEL_PROFILE	Which detection/segmentation pipeline to run	opencv-mvp or seg-v1