agentleFS
NVIDIAKnown publisherNVIDIA/IsaacTeleop/src/python/isaaccapture/cts/full_body/oracle/AGENTS.md

IsaacTeleop / oracle

AGENTS.md438 starsChanged 31 days agoApache-2.0View on GitHub

CRITICAL: complete the mandatory AGENTS.md preflight in ../../../../../../AGENTS.md before editing here, and read ../checker/AGENTS.md, which covers the thing this set judges. README.md says what each file is and how to build the set. Use these fixtures to prove a measurement is correct. Never to choose a threshold. Synthetic motion is smooth by construction (still windows are perfectly still, real ones carry tremor, foot sliding and tracker jitter) and injected faults are caricatures. Structural checks (up axis, units, handedness, quaternion order,…

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this AGENTS.md into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only the index is versioned, which is safe only because generation is byte-deterministic.

    line 34
  2. Very specific:

    Every seed is fixed (skeleton.build_clean, PostureParams.seed, every random.Random(...) in generate_fixtures.py).

    line 37
  3. Distinctive:

    Regenerate, never hand-edit: Add a defect as a new transformation in the generator so it composes with everything else.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md
<!-- SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. SPDX-License-Identifier: Apache-2.0 -->

Agent notes — cts/full_body/oracle/

CRITICAL: complete the mandatory AGENTS.md preflight in ../../../../../../AGENTS.md before editing here, and read ../checker/AGENTS.md, which covers the thing this set judges. README.md says what each file is and how to build the set.

Purpose and limits

Use these fixtures to prove a measurement is correct. Never to choose a threshold. Synthetic motion is smooth by construction (still windows are perfectly still, real ones carry tremor, foot sliding and tracker jitter) and injected faults are caricatures. Structural checks (up axis, units, handedness, quaternion order, joint indexing) need no threshold and are settled here; everything expressed as a tolerance is not. Do not treat a property the generator happens to produce as a property of real recordings.

Share no code with the checker

The 24-joint table, the parent table, the forward kinematics and the quaternion arithmetic exist twice on purpose: here in skeleton.py, and in ../checker/src/full_body_cts/profile.py and ../../common/src/cts_common/vectors.py. A fault in code shared by the oracle and the thing it judges cancels itself out. Do not resolve the duplication. ../checker/tests/test_boundaries.py::test_the_oracle_shares_no_code_with_the_checker asserts it.

The index is committed; the recordings are not

Only the index is versioned, which is safe only because generation is byte-deterministic. Preserve that:

  • Every seed is fixed (skeleton.build_clean, PostureParams.seed, every random.Random(...) in generate_fixtures.py).
  • Nothing reads a wall clock; write_mcap takes every timestamp from the frame.
  • mcap is pinned in ../checker/requirements.txt: the index records size_bytes, and a writer with different chunking would move every one.

generate.sh fails if fixtures_index.json comes out dirty; find which of the above broke rather than committing the new index.

Rules

  • Regenerate, never hand-edit. Add a defect as a new transformation in the generator so it composes with everything else.
  • The index is never edited to make a checker test pass. Declare a disagreement in ../checker/tests/known_deviations.py with its reason.
  • Check names in the index are the vocabulary. Each defect fixture's expected_failing_check specifies a check's name; do not rename one without the index.
  • Never put a real recording in here: human motion data, and too large for version control.
  • Test data the checker needs for itself does not belong here. Adding a fixture rewrites the index, so build cases about the checker rather than the device in memory with ../checker/tests/synth.py.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.