facebookexperimental/triton/amd-att-trace
Claude Code and agent skill by facebookexperimental/triton
- community
Checking that the SKILL.md file is still in the repository.
Collect, validate, package, and inspect rocprofv3 Advanced Thread Trace bundles for AMD GPU kernels. Use for ATT collection, viewer-ready artifacts, or loop cycle windows; do not trigger for ordinary performance benchmarks.
Install
npx vskill@latest install facebookexperimental/triton/amd-att-traceSKILL.md
AMD Advanced Thread Trace
Use scripts/att_trace.py for the reproducible parts of ATT collection. Do not recreate rocprofv3 commands or decide that a raw .att file is viewer-ready by inspection alone.
Authorization and scope
- Collect a new trace only when the user explicitly asks to run or profile a
workload. Reading, validating, or packaging an existing trace is read-only.
- Check GPU occupancy before collection and choose an idle device. Do not stop
another user's job or run killgpu.sh unless explicitly authorized.
- Keep ATT diagnostics separate from correctness and end-to-end performance
measurements. A trace is not a correctness result or a stable benchmark.
- Never overwrite an existing non-empty output directory or archive.
Workflow
- Establish the application command, kernel regex, steady-state dispatch
number, GPU, and output label. State any inferred dispatch numbering.
- Run
att_trace.py probe. Use a profiler that exposes--attand a matching
decoder directory. Prefer a profiler and target application from compatible ROCm installations.
- Use
collect --dry-runwhen command selection or quoting is uncertain. - Collect exactly the requested dispatch. Use a unique output directory.
- Treat collection as successful only when the automatic viewer-bundle
validation passes.
- Package the complete trace root with
package; never transfer only the raw
.att files.
- Report the trace root, archive and checksum if packaged, selected profiler
and decoder, kernel regex, dispatch number, and decoded UI directory.
Commands
Iteration windows are zero-based by default and end at the next loop-marker occurrence after --last. Choose a loop marker that appears once per loop iteration; barriers can produce multiple ATT events. Record unrolling such as "one outer iteration represents two logical K tiles" instead of silently equating outer-loop and logical-tile iterations.
References
- Read references/rocprofv3.md when choosing ATT
options, a profiler installation, or a dispatch.
- Read references/viewer-format.md when
validating decoded output, transferring a bundle, or deriving cycle windows.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingthird_party/tlx/tools/agents/skills/amd-att-trace/SKILL.md
Popularity
- Repository stars
- 204
- Repository forks
- 84
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/facebookexperimental/triton/amd-att-trace)