agentsclimarketplace

Tracking objects

Skill graph-robots/open-robot-skills/skills/tracking-objects

Skill and tool bundles for gap (graph as policy) — Anthropic Agent Skills format, discovered by path

Install
npx -y skills add graph-robots/open-robot-skills --skill tracking-objects

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Long-running skill that drives the SAM3 tracker from the graph-scoped observation stream. Seeds the tracker via text prompt on the first frame, then polls the stream at update_hz and advances via sam3.tracker_update until the workflow signals termination, publishing a tracker snapshot per tick. Use when a workflow needs the live mask + box of an object across many frames — e.g. a supervisor branch that monitors a target's location while a policy manipulates it.

SKILL.md

2.7 KB, as published. Nobody here has run it

tracking-objects

Long-running tracker skill. Init on first frame, update per tick, close on exit. Used as a parallel sibling to other long-running skills (e.g. a policy) when continuous state estimation is needed.

The skill is class-based and stateful: the tracker session id and the last good mask/box live on the skill instance, so repeated visits to the same state within one workflow execution can resume the session instead of re-seeding it (pass close_on_exit: false to keep the session open across visits; the final visit — or the instance teardown — closes it).

It is also a streaming skill (gap.streaming: true): each update tick publishes a tracker snapshot {mask, box, confidence, object_present, n_updates} via ctx.publish, so downstream {"$ref": "<node>"} consumers see the latest tracked state while the loop is still running.

Install

Depends on the sam3 tool bundle:

uv sync --extra sam3   # (pip: pip install -e "open-robot-skills[sam3]")

When to use

  • A workflow that needs the live mask + box of an object across many frames (e.g., a supervisor that monitors a target's location while a policy manipulates it).
  • Wrapped under a parallel state with a join_policy so the tracker is cooperatively cancelled when the sibling branch finishes (the loop checks ctx.cancel_token every tick).

Output

Returns the final mask, box, confidence, and a flag indicating whether the object was visibly present at exit. Intermediate updates are published as streaming snapshots; the return value exposes only the final state.

Tool form

The bundle also exposes the loop as a flat tool — tracking-objects.track — for callers that want to invoke it as a single unit (one fresh tracker session per call) rather than as a workflow state.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.