Menu
Diagnose a failing flyprof command by its error code. Use when any flyprof command returns ok:false / a non-zero exit, to decide whether to stop, fix config, retry, or treat the result as a finding. Maps each error.code to a hard action so you don't loop.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Analyze a captured FlyDSL kernel bundle with flyprof: read the roofline/occupancy (counters), tile advice, stall/bubble taxonomy (bubbles), and the ISA<->Python mapping (map). Use after `flyprof capture`, when you need to understand where a kernel spends its stall cycles and how the ISA lines up with the .py source.
Capture a rocprofv3 ATT (Advanced Thread Trace) for a FlyDSL kernel with flyprof, including kernel discovery, tile sizing, and handling empty/out-of-range dispatches. Use when you need to record a kernel's per-instruction trace before analysis, or when `flyprof capture` returned EMPTY_ATT / DISPATCH_OUT_OF_RANGE / NO_KERNEL_DISCOVERED.
Produce and act on a FlyDSL kernel optimization report with flyprof: synthesize the ranked bubble->FlyDSL-knob recommendations, read the ISA<->Python evidence, optionally apply one knob, and verify the change re-profiles better. Use when asked what to change to make a kernel faster, or to turn a captured bundle into an optimization plan.
Start here to profile a FlyDSL kernel on AMD CDNA (gfx950/MI350X). Router for the flyprof CLI: how to run the fixed capture->analyze->report workflow, read its JSON envelopes, and pick the right sub-skill. Use whenever asked to profile/analyze a kernel, find why a kernel is slow, build an optimization report, or capture an ATT trace for a FlyDSL kernel.
| name | flyprof-triage |
| description | Diagnose a failing flyprof command by its error code. Use when any flyprof command returns ok:false / a non-zero exit, to decide whether to stop, fix config, retry, or treat the result as a finding. Maps each error.code to a hard action so you don't loop. |
| allowed-tools | Bash(flyprof:*), Read |
Every failure is one envelope: {"ok": false, "error": {"code", "exitCode", "message", "help", ...}}. Read the code, take the bound action. Do not retry the same command
hoping for a different result.
| code | exit | action |
|---|---|---|
ROCPROFV3_MISSING | 69 | STOP. rocprofv3 isn't on PATH. Tell the user to install ROCm / fix PATH. Do not retry. |
NO_GPU | 69 | STOP. No GPU visible. Do not retry. |
BUILD_TREE_MISSING | 78 | STOP. No built FlyDSL. Fix --worktree to a checkout with build-fly/python_packages. |
BAD_ARGS | 78 | Fix the argument the message names (e.g. --shape M,N,dtype), then re-run. |
MISSING_PREREQ | 78 | Run the prerequisite command (error.help says which) into the same --bundle, then re-run. |
KERNEL_NOT_FOUND | 66 | The kernel isn't in the worktree. flyprof list to find the right name. Not a bug. |
NO_KERNEL_DISCOVERED | 66 | --stats saw no FlyDSL kernel. The test may not launch one; check error.candidates. Try a real --shape. |
EMPTY_ATT | 66 | All dispatches were empty shells — the grid was too small. Enlarge --shape. A result, not a crash. |
DISPATCH_OUT_OF_RANGE | 66 | Use a dispatch id from error.available. |
NO_ARTIFACT / NO_RECOMMENDATION | 66 | Empty ≠ broken. Likely a valid "nothing here" / "already optimal". Investigate, don't "fix". |
CAPTURE_TIMEOUT | 75 | Transient. Retry once with a bigger --timeout or a smaller --shape. Budget one retry. |
INTERNAL | 70 | A real bug in flyprof. Capture error.trace, report it. Don't loop. |
If capture reports the kernel failed to compile (e.g. a missing fast-dispatch slot),
that is a real result to report, not a signal to retry with tweaks. Surface it and stop.
At most 3 triage rounds for a single goal. If you've fixed config (worktree, shape, prerequisite) twice and still fail, stop and report the last envelope verbatim — don't keep mutating arguments.
