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 agentmods add commands/dogacel/auto-gpu-kernel/log-experimentgit clone --depth 1 https://github.com/Dogacel/auto-gpu-kernelWhat 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 | $0.00000 | $0.00427 |
| Opus 5 | $0.00000 | $0.00214 |
| Sonnet 5 | $0.00000 | $0.00085 |
| Haiku 4.5 | $0.00000 | $0.00043 |
Grade A, and why
log-experiment 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 2d 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.
What it actually says
/log-experiment
Log the most recent experiment. Never skip — failures are as valuable as wins.
Pick folder
List experiments/exp_*/. Let N = highest number.
- If
exp_N/plan.mdexists withoutresult.md→ useexp_N/. - Else → create
exp_(N+1)/. - No folders yet →
exp_1/.
Never overwrite an existing result.md. If you'd have to, stop and ask the user.
Write artifacts
- Copy
solution/triton/sparse_fused.pyinto the folder (same filename). - Copy the Modal log produced by
/benchmarktobench.login the folder. - Write
result.md:
# Experiment N — YYYY-MM-DD
**Description:** what changed, why. Reference `plan.md` when implementing one.
## Results
- Pass: X/Y
- Kernel latency (ms): small=S.SSS / large=L.LLL / overall=O.OOO (min / median / max)
- Reference latency (ms): R.RRR
- Max abs err: X.XXe-X | Max rel err: X.XXe-X
- Mode: quick | stride N | full (| ab-vs-exp_K if A/B)
## Learnings
What was learned. What to try or avoid next. If durable cross-experiment insight, also append one line to `experiments/LESSONS.md`.
- Append to
experiments/summary.md(create with header row if missing):
| Exp | Date | Description | Latency | Ref | Pass | Notes |
|---|---|---|---|---|---|---|
| N | YYYY-MM-DD | one phrase | O.OOO ms | R.RRR ms | X/Y | Δ% vs prior best, "new best" / "regression" / "ablation" |
Keep Notes terse. Detail lives in result.md.
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.
- 2d ago First seen · 45 lines · 0 tokens per session scan A 6a43de3f6af3
log-experiment is a command published in the GitHub repository Dogacel/auto-gpu-kernel (157 stars, last pushed 11d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 427 tokens. 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 2026-08-30.
Other commands, from other repositories
learn
Force claude-smart to extract learnings from this session now.
component
Scaffold a new React component grounded in the paper-mono primitives. Requires explicit kind or a nearest-existing-component match. No empty divs, no speculative scaffolding.
memory-store
Store an insight, decision, or pattern to memory.
review
Cold re-quiz on code that already shipped — your own session commits, not the change in front of you.
no-vibe
Enter no-vibe mode in OpenCode (tutor mode, no direct project file writes).
setup-content
Lesson command — 教材コンテンツの初回セットアップ.