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 skills/darwin-agent/car-bench-trace/sunroofnpx skills add Darwin-Agent/Car-bench-TRACE --skill sunroofgit clone --depth 1 https://github.com/Darwin-Agent/Car-bench-TRACEWrote 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/darwin-agent/car-bench-trace/sunroof)<a href="https://agentmods.dev/skills/darwin-agent/car-bench-trace/sunroof"><img src="https://agentmods.dev/badge/skills/darwin-agent/car-bench-trace/sunroof.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00000 | $0.02577 |
| Opus 5 | $0.00000 | $0.01288 |
| Sonnet 5 | $0.00000 | $0.00515 |
| Haiku 4.5 | $0.00000 | $0.00258 |
Grade A, and why
sunroof 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 4d 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 — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Open / close / position the sunroof
Open, close, or set the sunroof to an opening percentage. Requests range from a fully specified open to ones with an unstated percentage, a weather-conditional open, the mandatory sunshade-open precondition, or a missing control. The same method handles all of these.
When this applies
- "Open the sunroof to N%" / "close the sunroof" — a fully specified move.
- "Open the sunroof for some fresh air" — no percentage given.
- "Open the sunroof if it's warm enough / unless it's raining" — a conditional open.
- Any sunroof open, which carries the sunshade-fully-open precondition and the weather-confirmation gate.
Tools
get_user_preferences()— read the default opening percentage only when the percentage is meant to come from a standing preference.get_sunroof_and_sunshade_position()— read the current sunroof and sunshade positions; always read this before any open to learn whether the sunshade precondition is already met. A field may read"unknown"— unverifiable, not a value.get_weather(...)— read weather for the current location at the current day/time. Always read it before any open: it both decides the weather-confirmation gate (see Method step 3) and evaluates any user "if warm/unless raining" condition. Pass only the documented arguments (location id, month, day, hour, optional minutes) — adding an unsupported argument likeyearmakes the call error and forces a wasteful retry.open_close_sunroof({percentage})— set the sunroof opening percentage;0is fully closed. May be missing in this session.open_close_sunshade({percentage})— used only to satisfy/name the sunshade-open precondition below. The full sunshade method lives in thesunshadeskill; do not re-teach it here.
Method
- Before opening, gather state once, in parallel:
get_sunroof_and_sunshade_position()(always — to learn the sunshade state) andget_weather(<current location, current day/time>)(always — for the confirmation gate and any user condition). If the percentage is preference-driven, fetchget_user_preferences()in the same batch. Closing the sunroof needs none of this — just close it. - Apply the weather-confirmation gate. If the weather condition is not one of sunny, cloudy, or partly_cloudy (e.g. rain, snow, fog, thunderstorm, hail), opening the sunroof requires the user's explicit expressive "yes" first. State the condition, lay out exactly what you'll do (open the sunshade fully, then the sunroof to N%), and wait for confirmation — do not open on this turn. If the weather is clear, no weather confirmation is needed; proceed. This gate is the most common reason an open must pause, so never skip the weather read.
- Honour the hard precondition, in order: the sunshade must be fully open before the sunroof opens. Once cleared to act, bring the sunshade fully open (to 100%) first (the
sunshadeskill owns how), then open the sunroof. The user implicitly accepts this prerequisite — perform it as part of fulfilling the request; don't ask about it separately. Issue the two opens in sequence, sunshade then sunroof, not in parallel. - Act on the sunroof, with the correct argument. Open vs. close, and the right
percentagevalue — don't close when an open was asked, and don't omit the value.
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.
- 4d ago First seen · 84 lines · 0 tokens per session scan A bacd3a332c80
sunroof is a skill published in the GitHub repository Darwin-Agent/Car-bench-TRACE (7 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,577 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 2026-08-31.
Other skills, from other repositories
integrated-browser
Use this when working on the VS Code integrated browser ("browserView") to understand its architecture and mental model. Covers the embedded Chromium browser, its editor tab, navigation, overlay/layout, sessions, and agent browser tools under src/vs/platform/browserView and src/vs/workbench/contrib/browserView.
project-snmp-profiles-authoring
Use when editing Netdata SNMP profile YAMLs, topology SNMP profiles, ddsnmp profile parsing, or profile-format documentation. Requires checking source MIB field accessibility, especially MAX-ACCESS not-accessible INDEX objects, before adding or changing profile symbols.
hyperpod-version-checker
Check and compare software component versions on SageMaker HyperPod cluster nodes - NVIDIA drivers, CUDA toolkit, cuDNN, NCCL, EFA, AWS OFI NCCL, GDRCopy, MPI, Neuron SDK (Trainium/Inferentia), Python, and PyTorch. Use when checking component versions, verifying CUDA/driver compatibility, detecting version mismatches…
unifi-protect
How to manage UniFi Protect cameras and NVR — view cameras, smart detections, Find Anything detection search, recordings, snapshots, lights, sensors, Known Faces, license plates, and the Alarm Manager. Use this skill when the user mentions UniFi cameras, security cameras, NVR, recordings, motion detection, person…
pcbway
PCBWay PCB fabrication and assembly — turnkey/consigned assembly, design rules, ordering workflow. Alternative to JLCPCB for manufacturing. Use with KiCad. Use this skill when the user mentions PCBWay, needs turnkey assembly (PCBWay sources parts by MPN), has parts not available on LCSC, needs assembled boards with…
speaker-recognizer
Self-enroll a voice for speaker recognition. Activate ONLY when (a) a mic Unknown Speaker: turn carries a clear self-introduction ("I'm X", "my name is X", "tôi là X"...), (b) prior same-tag [voice:voiceN] turns can be combined with a now-named turn, (c) a Telegram voice-note carries an intro, OR (d) the user asks to…