vision-delivery

Run an offline Sentinel package doctor and separate local installation defects from host enablement or Roboflow sign-in problems. TRIGGER when: the user asks to check, verify, diagnose, or troubleshoot a Sentinel plugin installation; run the setup doctor; find missing skills, packaged files, or Roboflow tools; or confirm Sentinel installed correctly. SKIP when: Sentinel and Roboflow work and the user wants to count boxes or solve a CV task; the package is healthy but OAuth/sign-in alone failed (auth-setup); or the request is deployment economics/crossover only (estimate-economics).

Turn an independently evaluated CV model or pipeline into a verified delivery handoff without copying volatile platform recipes. TRIGGER when: user has a working model, Workflow, OCR pipeline, or tracker and asks to export it, run locally, integrate an app/camera/RTSP source, deploy, productionize, monitor drift, add active learning, or make a runnable package. SKIP when: user asks to build a detector/model from sample images, the capability is unsolved, or measured acceptance is absent (solve-cv-task or modality skill); the request is cost/crossover only (estimate-economics); it asks only for a current Roboflow API field, UI path, model ID, or platform how-to (official Roboflow resource); or it only post-processes existing masks/data.

Connect or troubleshoot Sentinel's URL-only Roboflow MCP sign-in. TRIGGER when: the user asks to connect Roboflow, sign in, authorize OAuth, configure the plugin, or diagnose unavailable/unauthorized Roboflow MCP tools; the user asks whether a Roboflow API key is required for plugin installation. SKIP when: Roboflow MCP tools already work; the task is purely local, such as discussing a labeling approach for raw footage, and needs no Roboflow operation; a generated standalone client needs credentials (delegate its exact auth shape to the current official Roboflow API/inference skill).

Produce one whole-image verdict: pass/fail, category, compliance, or anomaly. TRIGGER when: user asks whether an image is defective/compliant; to flag bad parts, classify product images, run image-level quality control, detect an anomaly, or build binary/multi-class classification. SKIP when: user asks how many boxes/pallets or for other object counts or per-person PPE (detect-and-analyze); masks/area (segment-and-analyze); identity across video (track-and-count); OCR/codes (read-text); to compare self-hosting vs managed or cost only (estimate-economics); or delivery/export of a working classifier (deliver-cv-project).

Generate a portable stakeholder decision report for a CV deployment decision. TRIGGER when: user explicitly asks for a decision report, manager report, or stakeholder doc after a cost crossover has been computed; phrases: "write me a decision report", "generate a report for my manager", "I need a stakeholder doc", "produce the decision doc", "decision-report skill", "report for my team". SKIP when: no cost crossover exists yet — estimate-economics must run first and deliver a crossover number; user is still in build work; user wants a one-paragraph summary (answer inline instead).

Decompose a complex vision-and-reasoning requirement into the cheapest measurable pipeline. TRIGGER when: user asks to monitor/alert, analyze footage over time, combine perception with aggregation/reporting, determine whether something is feasible to detect (including sensor limits or smoldering), or replace repeated vision-language calls with a cheaper system. SKIP when: the request is single-frame detection/counting (detect-and-analyze), whole-image classification (classify-or-flag), deployment cost only (estimate-economics), or the pipeline is already working and needs delivery (deliver-cv-project).

Detect instances and analyze boxes, counts, crops, regions, or metadata. TRIGGER when: user asks to count/detect objects, report the number of defects, build a model that counts trucks, tally boxes/pallets on a conveyor, find workers lacking PPE, crop detections, extract an ROI, estimate box-relative size, or report per-object confidence/zone metadata. SKIP when: user needs one whole-image verdict (classify-or-flag); identity/path/dwell/line crossing (track-and-count); masks/contours/calibrated area/crack width (segment-and-analyze); OCR/serials (read-text); to compare self-hosting vs managed or cost only (estimate-economics); active learning/export/delivery of a working model (deliver-cv-project).

Estimate CV annotation, training, deployment, and operations economics. TRIGGER when: user invokes `$estimate-economics` or `/sentinel:estimate-economics`, asks labeling/training cost, managed vs self-hosted, build-vs-buy, scale economics, deployment crossover, or selects managed-at-scale after a passing proof. SKIP when: user asks to build/test a detector/classifier/OCR/tracker/pose/mask, count objects, finish the PoC, or keep improving because the eval has not passed yet (solve-cv-task); asks only a current platform how-to; or lacks passing evidence unless explicitly accepting a rough estimate.

Extract structured text, numbers, labels, and codes from images and evaluate by character or field accuracy. TRIGGER when: user asks to read/scan a serial, part/date/lot number, label, license plate, meter, gauge, barcode, QR code, invoice, document, form field, or other image text. SKIP when: user asks to tally/count boxes or other objects on a conveyor (detect-and-analyze), pass/fail without text (classify-or-flag), measure crack width/masks/area (segment-and-analyze), track objects (track-and-count), cost only (estimate-economics), or integrate/deliver working OCR (deliver-cv-project).

Build keypoint-based pose, gesture, action, ergonomics, posture, or fall analysis. TRIGGER when: user asks for body/hand pose, gestures, action recognition, skeletons, keypoints, joint angles, raised hands, bending, squats, falls, sign language, or posture compliance. SKIP when: user needs an exact body contour/mask (segment-and-analyze), per-person PPE without keypoints (detect-and-analyze), whole-image compliance (classify-or-flag), identity/path tracking without pose (track-and-count), OCR (read-text), to compare cloud-hosted vs self-hosted or cost only (estimate-economics), or integration/delivery of a working pose model (deliver-cv-project).

Produce pixel-precise masks plus area, shape, perimeter, calibration, or physical measurements. TRIGGER when: user asks to segment/outline an object or defect; measure exact area, crack width, corrosion, lesion boundary, contours, millimeters, or calibrated dimensions; or quantify mask shape. SKIP when: user only counts boxes (detect-and-analyze), needs one image verdict (classify-or-flag), reads text (read-text), tracks identity (track-and-count), asks cost only (estimate-economics), already has a working mask model and needs delivery (deliver-cv-project), or only post-processes existing masks locally.

Turn a plain-language CV goal into the correct measurable build route. TRIGGER when: user asks to detect/count vehicles, read serial numbers, track people through zones, use a factory camera to catch mistakes, solve a CV problem from sample images, test feasibility, or build an end-to-end vision capability without knowing the modality. SKIP when: request clearly matches a specialist; asks managed deployment cost, annotation/training cost, or other economics only; asks current platform how-to; invokes estimate-economics; or says an evaluated model works and only needs package/export/deployment/delivery.

Track identity across video and produce paths, dwell, crossings, zones, or linked counts. TRIGGER when: user asks to follow a person/vehicle/object across frames, track shoppers/forklifts, count entries/line crossings, ask how many vehicles crossed an intersection, analyze video with identity, measure dwell/path, monitor RTSP, or alert on zone entry. SKIP when: user asks to count cars in a parking lot now or other per-image/per-frame counts (detect-and-analyze); masks/area (segment-and-analyze), one image verdict (classify-or-flag), OCR (read-text), cost only (estimate-economics), or export/delivery of a working tracker (deliver-cv-project).