Menu
Label visible cleanup-relevant movable objects from grouped RAW-FPV frames without creating executable cleanup handles.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Select and run Roboclaws validation, product, eval-suite, and live-agent eval rows from a plan, diff, or explicit capability request.
Complete open-ended household robot goals through public household MCP tools.
Drive the ADR-0003 MolmoSpaces real-world cleanup MCP contract.
Align scene Gaussian/splat, USD/mesh, and robot map assets into an honest digital-twin evidence workflow. Use when a new Gaussian scene arrives, when B1/Map12-style assets need to be connected, when map anchors are being projected into a 3D scene, or when an agent must decide whether an alignment is candidate, verified, runtime-proven, or planner-backed.
Create blog, README, social, or demo-review visual showcases from completed Roboclaws run artifacts, especially chaptered GIFs/contact sheets for household cleanup runs. Use when asked to turn report output, robot_views, traces, or run_result.json into polished visual result evidence without rerunning the robot task.
Use this skill whenever the user asks whether a Roboclaws live-agent report run is faster, wants to compare Agent SDK/Codex/Claude cleanup reports, asks about model-call work, residual latency, speedup validity, or report performance under provider/network variance. The skill extracts sanitized JSON metrics first and answers only from generated metric/comparison artifacts, not from report.html scraping or raw prompts.
| name | raw-fpv-visual-labeler |
| description | Label visible cleanup-relevant movable objects from grouped RAW-FPV frames without creating executable cleanup handles. |
Use this skill only for perception-only RAW-FPV labeling probes. It consumes
public robot FPV frame evidence and emits structured visual labels for review
and offline scoring. It does not call cleanup tools, does not create
observed_* handles, and does not authorize navigation, pick, place, or done.
Use 3-6 neighboring RAW-FPV frames from the same waypoint, sweep segment, or source observation neighborhood when available. Each frame may include only:
Never include private labels, generated hidden target ids, acceptable destination truth, executable observed-object handles, detector candidates, or camera-label producer candidates.
Return strict JSON:
{
"schema": "raw_fpv_visual_labeler_response_v1",
"labels": [
{
"evidence_frame_id": "run/raw_fpv_001",
"category": "mug",
"category_family": "dish",
"coarse_region": "middle_right",
"confidence": 0.82,
"is_cleanup_relevant": true,
"bbox": [0.62, 0.5, 0.12, 0.15],
"surface_hint": "table",
"reason_not_actionable": ""
}
]
}
Required per label:
evidence_frame_idcategorycategory_familycoarse_regionconfidenceis_cleanup_relevantOptional per label:
bboxsurface_hintreason_not_actionableAllowed category families are food, dish, book, linen, toy, and
electronics. Fixtures and surfaces such as tables, beds, counters, shelves,
sinks, cabinets, and floors may be mentioned only as surface_hint or as
is_cleanup_relevant=false; they are not object hits.
These labels are perception evidence. The cleanup agent must not consume them
as executable handles in camera-raw-fpv. A later assisted RAW-FPV or
camera-grounded-labels producer decision would need its own contract change.