agentleFS
AnthonyPoschenAnthonyPoschen/agent-skills/skills/renderdoc-gpu-debug/SKILL.md

renderdoc-gpu-debug

Skill2 starsChanged 21 days agoMITView on GitHub

Use when capturing or inspecting a Vulkan frame with RenderDoc on zanven-pc (RTX 4070): .rdc captures, GLSL or SPIR-V shader debugging, pipeline state, or pixel history via rdc-cli. Do not use for CSS, HTML, canvas, or browser rendering.

  • Installs packages

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the current host is not zanven-pc, do not run rdc, renderdoccmd, or any capture script.

    line 27
  2. Very specific:

    Use --session name when two captures must stay open.

    line 58
  3. Distinctive:

    Run the app from the directory it uses to find assets:

    line 62
  4. Distinctive:

    Start small. Full event lists are too large.

    line 77
  5. Distinctive:

    Do not dump image bytes through the shell.

    line 105
  6. Very specific:

    Prefer the default table over --json until the set is small.

    line 135
  7. Very specific:

    Ask for one pipeline section (rdc pipeline EID ps), not the full state.

    line 136

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

RenderDoc GPU Debugging

Inspect Vulkan frame captures with host rdc-cli (RenderDoc's Python API). D3D11, D3D12, and OpenGL captures use the same commands. This skill does not include the upstream MCP server.

Source and MIT licence: LICENSE-UPSTREAM.

Where capture and replay run

Capture and replay run only on Anthony's GPU machine zanven-pc (RTX 4070), which has host RenderDoc and rdc-cli. Cloud VMs and the bot box have no GPU.

If the current host is not zanven-pc, do not run rdc, renderdoccmd, or any capture script. Hand the capture and replay steps to a session on zanven-pc. Opening an existing .rdc is replay: it also belongs on that machine.

Prerequisites

On zanven-pc, rdc doctor must pass before capture or inspection.

  • rdc-cli is installed (pip install rdc-cli==0.7.0 if the command is missing).
  • RENDERDOC_PYTHON_PATH points at the RenderDoc Python module built for the host Python. On Linux that module is renderdoc.so. On Windows it is renderdoc.pyd, and renderdoc.dll sits in the same directory. The Windows names do not apply on Linux.
  • renderdoccmd is on PATH (Windows binary name: renderdoccmd.exe).

Vulkan needs the RenderDoc implicit layer and ENABLE_VULKAN_RENDERDOC_CAPTURE=1. On Linux, point the loader at the layer JSON with VK_ADD_LAYER_PATH or the user implicit_layer.d directory. The Windows registry key HKCU\SOFTWARE\Khronos\Vulkan\ImplicitLayers applies only on Windows.

Session

Open one capture, inspect, then close. A leaked daemon holds GPU memory.

rdc open path/to/capture.rdc
rdc status
rdc close

Use --session name when two captures must stay open.

Capture

Run the app from the directory it uses to find assets:

cd /path/to/app && rdc capture -o capture.rdc -- ./app [app-args...]

Useful flags: --frame N, --timeout S (default 60), --api-validation, --ref-all-resources, --wait-for-exit.

RenderDoc frame boundaries are present calls. A headless Vulkan app with no swapchain cannot be captured. On zanven-pc, prefer an interactive present, then --trigger plus rdc capture-trigger, then renderdoccmd capture.

Frame exploration

Start small. Full event lists are too large.

rdc info --json
rdc stats --json
rdc passes
rdc draws --limit 20
rdc draws --pass "Shadow Pass" --json
rdc events --limit 50
rdc event 42 --json
rdc ls /textures -l

Pipeline, shaders, and pixels

rdc pipeline EID --json          # or one section: vs ps om rs ia ds
rdc bindings EID --set 0 --json
rdc shader EID ps --source
rdc shader EID ps --target spirv
rdc shader EID ps --constants --json
rdc pixel X Y EID --json
rdc pick-pixel X Y EID --json
rdc debug pixel EID X Y --trace
rdc debug vertex EID VTXID --json

Export images next to the capture, then view the PNG with the image read tool. Do not dump image bytes through the shell.

rdc rt EID -o captures/render_target.png
rdc texture RESID --mip 2 -o captures/mip2.png

Shader edit and replay, still on zanven-pc:

rdc shader-encodings --json
rdc shader EID ps --source -o captures/shader.frag
rdc shader-build captures/shader.frag --encoding GLSL --stage ps --json
rdc shader-replace EID ps --with SHADER_ID --json
rdc rt EID -o captures/after_edit.png
rdc shader-restore EID ps

Compare two captures:

rdc diff before.rdc after.rdc --shortstat
rdc diff before.rdc after.rdc --framebuffer --diff-output captures/diff.png

Output size

  1. rdc draws --limit 20 and rdc events --limit 50 while exploring.
  2. Filter with --pass before listing every draw.
  3. rdc draws -q returns event ids only.
  4. Prefer the default table over --json until the set is small.
  5. Ask for one pipeline section (rdc pipeline EID ps), not the full state.

When a command fails

SymptomWhat to do on zanven-pc
rdc not foundpip install rdc-cli==0.7.0
rdc doctor failsFix RENDERDOC_PYTHON_PATH and the host RenderDoc module
Capture reports no swapchainrdc capture --trigger, then rdc capture-trigger
Daemon not respondingrdc status, rdc close, rdc open again
Counters or pixel history emptyrdc gpus --json and rdc counters --list; the driver may omit them

References

Command flags and defaults: commands-quick-ref.

Pass, colour, shadow, and pixel recipes: debugging-recipes.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.