Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add OutlineDriven/outline-driven-development --skill cuda-debugginggit clone --depth 1 https://github.com/OutlineDriven/outline-driven-developmentWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/outlinedriven/outline-driven-development/cuda-debugging)<a href="https://agentmods.dev/skills/outlinedriven/outline-driven-development/cuda-debugging"><img src="https://agentmods.dev/badge/skills/outlinedriven/outline-driven-development/cuda-debugging/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/outlinedriven/outline-driven-development/cuda-debugging"><img src="https://agentmods.dev/badge/skills/outlinedriven/outline-driven-development/cuda-debugging.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00054 | $0.01691 |
| Opus 5 | $0.00027 | $0.00846 |
| Sonnet 5 | $0.00011 | $0.00338 |
| Haiku 4.5 | $0.00005 | $0.00169 |
Grade A, and why
cuda-debugging scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 136 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CUDA debugging
Contract
| Field | Bound contract |
|---|---|
| Trigger | A CUDA program crashes, hangs, or returns wrong values, and the job is interactive GPU thread inspection, automated memory and race checks, core dump analysis, device printf, or runtime error triage. |
| Authority | Reversible local. Writes are limited to debug builds, sanitizer logs, and core dumps in the project tree or /tmp; rollback is deleting those files and rebuilds. No remote mutation. |
| Side effect | Rebuilt binaries, sanitizer reports, core dump files, and a triage verdict. |
| Done | The defect class is named with evidence: a sanitizer report, a thread-level observation, or an error-code diagnosis, and the fix location is identified. |
Inputs
- Symptom (required): crash, wrong values, hang, or timeout, and when it appears.
- Kernel source (required): the
.cufiles under suspicion. - Toolkit (required):
nvcc --versionandcompute-sanitizer --version. Grounded current stable: CUDA Toolkit 13.3 Update 1. - Target GPU and any MIG layout (optional):
nvidia-smi -L.
Procedure
- Build for the tool. Source-level stepping in cuda-gdb needs
nvcc -G -g -O0. Sanitizers accept optimized builds and report clearer lines with-lineinfo -g -O2:
nvcc -G -g -O0 -arch=sm_80 -o app_debug main.cu
nvcc -lineinfo -g -O2 -arch=sm_80 -o app_san main.cu
Done when: both binaries exist and -G is absent from the sanitizer build.
- Run the Compute Sanitizer tools in order of suspicion. Each tool slows execution sharply; run one at a time:
compute-sanitizer --tool memcheck ./app_san # out-of-bounds, misaligned, leaks
compute-sanitizer --tool racecheck ./app_san # shared and global memory races
compute-sanitizer --tool initcheck ./app_san # uninitialized reads
compute-sanitizer --tool synccheck ./app_san # missing or misaligned __syncthreads
compute-sanitizer --tool memcheck --log-file san.log ./app_san
A typical memcheck hit names the kernel, line, thread, and address:
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 136 lines · 54 tokens per session scan A 2231c20c5178
cuda-debugging is a skill published in the GitHub repository OutlineDriven/outline-driven-development (52 stars, last pushed 4d ago), licensed Apache-2.0. It adds 54 tokens to every session and 1,691 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-06.
Other skills, from other repositories
eagle3-triage
Triage a failed EAGLE3 pipeline run. Identifies which step failed (data synthesis, hidden state dump, training, or benchmark), diagnoses root cause from logs, and suggests fixes. Use when user reports an EAGLE3 pipeline failure or asks why a specific step failed. Also helps debug new model support issues.
trulens-instrumentation
Instrument LLM apps with TruLens OTEL-based tracing - from setup to debugging and optimization.
tracely
Instrument AI agents with Tracely and turn their production traces into CI gates. Use when the user mentions Tracely, tracely-ai, tracelysdk, the tracely CLI, or asks to trace/observe an AI agent, add LLM evaluators or LLM-as-a-judge columns, debug why a trace or conversation isn't showing up, wire agent regression…
profile-model
Measure one TensorRT-Model-Connect model through its public Task API or run a checked-in performance-matrix entry with comparable evidence.
troubleshooting-dbt-job-errors
Diagnoses dbt Cloud/platform job failures by analyzing run logs, querying the Admin API, reviewing git history, and investigating data issues. Use when a dbt Cloud/platform job fails and you need to diagnose the root cause, especially when error messages are unclear or when intermittent failures occur. Do not use for…
agent-observability-trace-rca
Root cause analysis on production LLM traces. Diagnoses why an LLM application is failing — works from eval judge verdicts, runtime errors, or structural anomalies depending on what signals are present. Walks the span tree from symptom to root cause. Use when user says "what's wrong with my app", "why is my eval…