OpenParable
Skip to content
← Back to skills

Debug Kernel

Testing & Quality
debug-kernel

Debug Kernel is an Agent Skill for testing & quality published in the tenstorrent/tt-metal repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Debug compilation or runtime errors in LLK kernels. Infers architecture and kernel type from path.

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/tenstorrent/tt-metal/blob/HEAD/tt_metal/tt-llk/.claude/skills/debug-kernel/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/debug-kernel, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

/debug-kernel — Debug an LLK Kernel

Usage

/debug-kernel tt_llk_blackhole/llk_lib/llk_pack_untilize.h
/debug-kernel tt_llk_quasar/common/inc/sfpu/ckernel_sfpu_gelu.h
/debug-kernel  (no args — will ask for the kernel path or check recent test output)

What to Do

  1. Parse the kernel path from the user's arguments. If not provided, check recent test output or ask.
  2. Infer the kernel type from the path:
    • common/inc/sfpu/ → sfpu
    • llk_lib/llk_math_ → math
    • llk_lib/llk_pack_ → pack
    • llk_lib/llk_unpack_ → unpack
  3. Infer the architecture from the path:
    • tt_llk_wormhole_b0/ → wormhole
    • tt_llk_blackhole/ → blackhole
    • tt_llk_quasar/ → quasar
  4. Spawn the llk-debugger agent:
Agent tool:
  subagent_type: "llk-debugger"
  description: "Debug {kernel_name}"
  prompt: |
    Kernel path: {kernel_path}
    Kernel type: {kernel_type}
    Architecture: {architecture}
    Error context: {any error output the user provided, or "check logs"}
  1. Report the agent's results to the user
  2. If the fix involves code changes, show the diff and explain the fix