Skip to content

agents.md

Latest commit

 

History

History
61 lines (44 loc) · 3.4 KB

agents.md

File metadata and controls

61 lines (44 loc) · 3.4 KB
Raw

Using SimLoop with an AI agent

SimLoop ships a toolkit for teams who want to program with an AI coding agent: a context file that teaches any agent how the library works, four skills for the jobs that come up, prompts to start from, and notes on spending less.

It lives in simloop-agents/ in the repository. Copy the folder into your FTC project.

The toolkit is optional Markdown. No build file reads it, and removing it does not affect the sim.

Before using it

Someone on the team needs to understand and review the code the agent changes. Ask it to explain the units, tick order, hardware names and assertions, then compare those with your robot. A passing simulation checks the model you supplied; it does not validate the physical robot.

The toolkit README has the setup instructions and suggested workflow.

Which tool

The instructions use the AGENTS.md format. Check how your coding tool loads project instructions and confirm that it has read this file. The skills and prompts are plain Markdown, so you can also provide them directly.

Horizon uses Claude Code; its paths are the ones the team has exercised most.

What is in the folder

File What it is
AGENTS.md The library overview, API inventory, simulation rules and test structure. Copy it to your repository root and add your robot's details.
skills/ Four procedures — write a sim test, add a mechanism, debug a failing sim, review sim code. Written for any agent; drop into .claude/skills/ unmodified.
prompts/ Copy-paste starting points, including one for learning and one for auditing your tests before a competition.
pointers/ One-line files for the tools that need a nudge, and for a project that already has its own instructions file. They point at AGENTS.md rather than copying it.
token-savers.md Ways to keep context focused and compare usage without assuming fixed savings.

The rules an agent gets wrong

These are what AGENTS.md exists to prevent. Each one produces code that compiles, and most of them produce a test that passes:

  • Inventing an API. A plausible method name half-remembered from another library. The surface is small and completely listed; what it does not do is the page to check first.
  • The log-key prefix. Logger.recordOutput("Slide/x", v) reads back as "RealOutputs/Slide/x".
  • Tick order. periodic() then advancePhysics(). Reversed lags every response one tick and still passes.
  • The two frames. Convert a pose when it crosses between plant and field; the wrong frame can produce incorrect geometry without throwing. See field.
  • The wall clock. Robot decisions must use sim time so repeated runs can be compared.
  • Assertions that cannot fail. Asserting a slide stayed within the end stops the plant clamps to every tick tests the model, not the robot.

Check that the assertion can detect a failure

Make every test fail once, on purpose, before you trust it.

Temporarily break the behaviour the test is meant to check: remove the command, use an unreachable target or shorten the run. Confirm that the relevant assertion fails, then restore the code. If the test still passes, it is not checking the behaviour you intended.