TheRock/rocm-systems/projects/amdsmi/CLAUDE.md

CLAUDE.md — AMD SMI Agent Guide

Project Overview

AMD SMI is a multi-language system management library for AMD GPUs and EPYC CPUs. Core C++ library (libamd_smi.so) with Python bindings, CLI, Go shim, and Rust bindings.

Critical Rules

1. Never edit py-interface/amdsmi_wrapper.py manually — it's auto-generated. Regenerate with tools/update_wrapper.sh or cmake -DBUILD_WRAPPER=ON.
2. PRs target develop branch (not main).
3. Pre-commit must pass before review: pip install pre-commit && pre-commit install
4. Version is defined in include/amd_smi/amdsmi.h (AMDSMI_LIB_VERSION_MAJOR/MINOR/RELEASE). CMake extracts it from there.
5. Excluded from formatting/linting: docs/, build/, esmi_ib_library/, third_party/, *.md, *.rst
6. Agent working files are ephemeral — write specs, plans, handoff docs, and scratch notes to ${TMPDIR:-/tmp}/amdsmi-agent-*, never into the workspace. The only curated agent docs that belong in git are .claude/context/CONTEXT.md and .claude/context/agent-flow.md.

Behavioral Guidelines

Bias toward caution over speed. For trivial tasks, use judgment.

1. Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs.

• State your assumptions explicitly. If uncertain, ask.
• If multiple interpretations exist, present them — don't pick silently.
• If a simpler approach exists, say so. Push back when warranted.
• If something is unclear, stop. Name what's confusing. Ask.

When asking the user to decide:

• Before surfacing a question, resolve what you can yourself (read the code, the docs, or dispatch a subagent). Only ask about things that genuinely need the user's judgment.
• Present choices as a lettered multiple-choice list (A / B / C) with a one-line tradeoff for each, and recommend one. Default to asking in chat; only write a /tmp doc if the user asks for one.

2. Simplicity First

Minimum code that solves the problem. Nothing speculative.

• No features beyond what was asked.
• No abstractions for single-use code.
• No "flexibility" or "configurability" that wasn't requested.
• No error handling for impossible scenarios.
• If you write 200 lines and it could be 50, rewrite it.

Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.

3. Surgical Changes

Touch only what you must. Clean up only your own mess.

When editing existing code:

• Don't "improve" adjacent code, comments, or formatting.
• Don't refactor things that aren't broken.
• Match existing style, even if you'd do it differently.
• If you notice unrelated dead code, mention it — don't delete it.

When your changes create orphans:

• Remove imports/variables/functions that YOUR changes made unused.
• Don't remove pre-existing dead code unless asked.

The test: Every changed line should trace directly to the user's request.

4. Goal-Driven Execution

Define success criteria. Loop until verified.

Transform tasks into verifiable goals:

• "Add validation" → "Write tests for invalid inputs, then make them pass"
• "Fix the bug" → "Write a test that reproduces it, then make it pass"
• "Refactor X" → "Ensure tests pass before and after"

For multi-step tasks, state a brief plan:

1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]

These guidelines are working if: fewer unnecessary changes in diffs, fewer rewrites due to overcomplication, and clarifying questions come before implementation rather than after mistakes.