AGENT SKILL

openpilot

Guides development, route/log analysis, car-porting, runtime services, simulator, and visual-debug workflows for comma.ai openpilot.

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

01 前言 Front Matter

name
openpilot
license
MIT
metadata
{ "disco-role": "operating" }
description
Guides development, route/log analysis, car-porting, runtime services, simulator, and visual-debug workflows for comma.ai openpilot.
disable-model-invocation
true

02 资源文件 4

03 SKILL.md 正文

openpilot Repo Skill

Use this operating skill when a task involves the comma.ai openpilot repository, package, developer tooling, route logs, car ports, controls, runtime services, simulator, or visualization/debug tools. openpilot is an open source driver-assistance/robotics system with Python, C/C++, Cython, Cap'n Proto, SCons, uv, and several submodule packages such as opendbc and msgq.

Read references/repo-provenance.md before treating this skill as current for a checkout. Read references/troubleshooting.md first when installs, imports, native extensions, submodules, or hardware assumptions fail.

First decisions

  1. Is the task about editing or validating the repo? Use development-and-testing.
  2. Is the task about route IDs, rlogs/qlogs, message filtering, alerts, or LogReader? Use route-log-analysis.
  3. Is the task about a car port, fingerprint, DBC/opendbc interface, controller behavior, process replay, or maneuver report? Use car-ports-and-controls.
  4. Is the task about cereal/msgq, Params, manager processes, loggerd/uploader/deleter, service timing, or hardware abstraction? Use core-services-and-runtime.
  5. Is the task about replay, Cabana, PlotJuggler/JotPluggler, simulator, joystick, UI, camera streams, or CTF-style tool exploration? Use simulator-and-visual-tools.

Safe baseline setup for a target checkout

In a target openpilot checkout, prefer the repository's uv/SCons workflow and Python 3.12 line:

# Inspect before mutating anything
python3 --version
git submodule status

# Typical developer setup from a fresh checkout
uv sync --frozen --extra tools --extra testing
scons -u openpilot/common/libparams_c.so msgq_repo/msgq/ipc_pyx.so msgq_repo/msgq/visionipc/visionipc_pyx.so

Use the bundled scripts/openpilot_skill_doctor.py or the development sub-skill's scripts/openpilot_dev_check.py to inspect a checkout without running mutable setup commands.

Important constraints

  • openpilot metadata requires Python >=3.12.3,<3.13.
  • Empty submodule directories are a common cause of install/import failure. Required submodules include msgq_repo, opendbc_repo, panda, rednose_repo, teleoprtc_repo, and tinygrad_repo.
  • Many imports need native outputs from SCons, especially openpilot/common/libparams_c.so, msgq_repo/msgq/ipc_pyx.so, and msgq_repo/msgq/visionipc/visionipc_pyx.so.
  • A CPU developer host can validate package imports, route/log parsing, unit-test collection, car docs, and many controller tests. It cannot validate in-car safety, Panda USB, AGNOS, camera/encoder power, GUI, simulator, or authenticated remote-route behavior by itself.
  • Treat live car/device commands, joystick control, op switch, updater/start/stop flows, and Params writes as state-mutating. Do not run them unless the user explicitly asks and prerequisites are clear.

Sub-skill map

Need Read
Setup, build, lint, tests, docs, CI-like checks development-and-testing/SKILL.md
Route/segment formats, LogReader, qlog/rlog summaries, filtering route-log-analysis/SKILL.md
Car porting, fingerprints, controls, process replay, maneuvers car-ports-and-controls/SKILL.md
cereal/msgq, Params, manager, loggerd, service diagnostics core-services-and-runtime/SKILL.md
Replay, Cabana, PlotJuggler, simulator, joystick, UI flags simulator-and-visual-tools/SKILL.md

Repo-level references and scripts

Verification posture

CPU import/build preparation was verified for the selected developer workflows. Hardware, remote-route network, GUI, and simulator capabilities are documented as optional or unverified unless the user provides the required device, route access, display, or dependency stack.

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