Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add graph-robots/open-robot-skills --skill perceiving-object-partsgit clone --depth 1 https://github.com/graph-robots/open-robot-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/graph-robots/open-robot-skills/perceiving-object-parts)<a href="https://agentmods.dev/skills/graph-robots/open-robot-skills/perceiving-object-parts"><img src="https://agentmods.dev/badge/skills/graph-robots/open-robot-skills/perceiving-object-parts/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/graph-robots/open-robot-skills/perceiving-object-parts"><img src="https://agentmods.dev/badge/skills/graph-robots/open-robot-skills/perceiving-object-parts.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00137 | $0.02040 |
| Opus 5 | $0.00068 | $0.01020 |
| Sonnet 5 | $0.00027 | $0.00408 |
| Haiku 4.5 | $0.00014 | $0.00204 |
Grade A, and why
perceiving-object-parts scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 191 lines — stays where its author put it; the contents beside it link to each section on GitHub.
perceiving-object-parts
Two-step zoom-in perception. The full image gives a small subpart (e.g. a frypan handle is ~3% of pixels) bad signal-to-noise for SAM3 text segmentation; cropping to the parent first brings the subpart up to ~30% of pixels in the cropped image — within SAM3's reliable range.
About
parent_promptandsubpart_prompt: they are literal Python strings, NOT subgraph inputs. They are author-time constants per subgraph instance. DO NOT declare them in the subgraph's top-levelinputsblock, and DO NOT writeRef("in.parent_prompt")or any otherRef(...)for them. Write the strings directly on the inner script node, e.g."parent_prompt": "frying pan", "subpart_prompt": "long horizontal handle of the frying pan". Onlycamerasis a flowed subgraph input (wired from the workflow's observation source, identical toperceiving-objects'scamerasinput).
When to use
- The grasp/place affordance is a part of a larger object (pan handle, drawer pull, moka-pot grip, mug rim, stove burner).
- Plain
perceiving-objectswithobject_name="handle"fails because there are multiple handles in the scene (drawer pull, microwave door, cabinet, ...) and DINO can't disambiguate.
When NOT to use
- The whole object IS the target (
perceiving-objectsis faster and produces a cleaner OBB). - The subpart spans the majority of the image already (skip the crop).
Pipeline
observation # rgb + depth + intrinsics + camera pose
│
▼ grounding-dino.detect(rgb, parent_prompt)
parent_box (BoundingBox2D) # broadest of the boxes, or VLM-picked
│
▼ crop_rgb_to_box(parent_box, padding=30)
cropped_image # H_new × W_new × 3 uint8
│
▼ grounding-dino.detect(crop, subpart_prompt) → sam3.segment_text
cropped_mask # subpart mask in crop coordinates
│
▼ uncrop(cropped_mask → original H × W)
full_mask # H × W uint8, zeros outside crop
│
▼ geometry.mask_to_world_points(full_mask, depth, K, T_cam)
world_cloud (PointCloud)
│
▼ geometry.filter_noise → geometry.compute_obb
subpart_obb # the split calls keep the unfiltered-cloud
# fallback when DBSCAN strips too many points
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 191 lines · 137 tokens per session scan A c4410d4a3a15
perceiving-object-parts is a skill published in the GitHub repository graph-robots/open-robot-skills (41 stars, last pushed today), licensed Apache-2.0. It adds 137 tokens to every session and 2,040 once invoked, about $0.0007 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
mjlab-to-mjswan
Port an mjlab task to mjswan: turn a local or remote GitHub repo that registers mjlab tasks into a browser app. Use when the user wants to run, visualize, or deploy an mjlab task (theirs or a third party's) in the browser with mjswan.
gap
Program robots with GaP (graph-as-policy) — compile natural-language tasks into typed, verified robot skill graphs and run them on simulators or real robots. Use when the user mentions GaP or graph-as-policy, robot manipulation, robot skills, robot tools or capabilities, skill registries, open-robot-skills, LIBERO or…
authoring-pipelines
Use when the user wants a bagel data pipeline — reducing a log to windows around events ("keep 30s around every hard brake"), recurring snippets/GIFs/exports, or any tasks-and-gates automation over robot data.
habitat-gs-train
Train and evaluate a navigation policy in the habitat-gs simulator. Covers the full generate-episodes → train → evaluate flow for PointNav / ImageNav / ObjectNav (Habitat-Lab + DDPPO reinforcement learning) and for Vision-and-Language Navigation (StreamVLN, Uni-NaVid). Use when the user wants to train, fine-tune…
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.