AGENT SKILL

task-authoring

Author RoboTwin tasks, task configs, language templates, and safe instruction-expansion workflows.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
task-authoring
license
MIT
metadata
{ "disco-role": "operating" }
description
Author RoboTwin tasks, task configs, language templates, and safe instruction-expansion workflows.
disable-model-invocation
true

02 资源文件 6

03 SKILL.md 正文

RoboTwin task authoring

Use this sub-skill when a user asks to add or modify a RoboTwin task, create a task configuration, edit language instructions, expand per-episode instructions, or understand the repository's LLM/code-generation utilities. If you only have the generated skill tree, use the root workspace bootstrapper first to materialize a pinned public workspace before editing runtime files.

Route first

  1. Work in the user's RoboTwin workspace. Runtime files here describe relative workspace files such as envs/<task_name>.py, description/task_instruction/<task_name>.json, and env_cfg/task_config/<config_name>.yml; do not edit this generated skill tree unless the user is updating the skill itself.
  2. Decide whether the request can be deterministic. Prefer manual edits plus the bundled safe scripts below. Do not invoke Azure, OpenAI-compatible, DeepSeek, Moonshot, or other hosted model APIs unless the user explicitly asks and provides credentials.
  3. For SAPIEN scene setup, Base_Task, actor/action helpers, robot/camera details, or render checks, route to simulation-core. For demonstration collection, HDF5 layout, downloads, and conversion, route to data-pipeline.

Main workflows

Guardrails

  • Keep canonical task implementations in envs/<task_name>.py; treat envs_gen/gpt_<task_name>.py as generated scratch code until manually reviewed and simulation-validated.
  • Keep play_once() placeholder output synchronized with instruction templates. If play_once() returns {"{A}": ..., "{a}": ...}, every non-arm placeholder required by a template must be present in that info mapping.
  • Do not assume RoboTwin is importable as an installed pip package. Many workflows expect commands to run from the RoboTwin workspace with local relative paths.
  • Asset setup matters: top-level envs imports can fail before assets/objects/objaverse/list.json and other assets are present, especially when cluttered-table utilities are imported.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351