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.
git clone --depth 1 https://github.com/vast-ai/vast-claude-pluginWrote 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/commands/vast-ai/vast-claude-plugin/search)<a href="https://agentmods.dev/commands/vast-ai/vast-claude-plugin/search"><img src="https://agentmods.dev/badge/commands/vast-ai/vast-claude-plugin/search.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 5.1 | $0.00014 | $0.00537 |
| Opus 5 | $0.00007 | $0.00269 |
| Sonnet 5 | $0.00003 | $0.00107 |
| Haiku 4.5 | $0.00001 | $0.00054 |
Grade A, and why
search 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 7d 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
/vastai:search
Search the Vast.ai marketplace for GPU offers and show the cheapest matches. Pass a filter as $ARGUMENTS using the same query syntax as vastai search offers.
Steps
-
Build the filter. If
$ARGUMENTSis empty, default to a renter-friendly query:'num_gpus=1 rentable=true verified=true'Otherwise pass
$ARGUMENTSthrough verbatim, ensuring it hasrentable=true(add it if missing — un-rentable offers are noise). -
Run the search, sorted cheapest-first:
vastai search offers <FILTER> -o 'dph_total' --raw -
Show the top 10 results as a table with: offer id,
gpu_name×num_gpus,dph_total($/hr),dlperf(perf score), andgeolocation. Don't print all 500+ rows — top 10 is enough for the user to pick from. -
If
$ARGUMENTSmentions a spot/bid intent ("spot", "bid", "cheapest interruptible"), use--type bidinstead of the default on-demand. -
If the result set is empty, suggest relaxing the filter (e.g., drop
verified=trueor widen the GPU type).
Examples
| Prompt | Filter | What runs |
|---|---|---|
/vastai:search |
default | vastai search offers 'num_gpus=1 rentable=true verified=true' -o 'dph_total' --raw |
/vastai:search gpu_name=RTX_4090 |
passthrough | vastai search offers 'gpu_name=RTX_4090 rentable=true' -o 'dph_total' --raw |
/vastai:search gpu_name=H100 num_gpus>=4 |
multi-GPU | vastai search offers 'gpu_name=H100 num_gpus>=4 rentable=true' -o 'dph_total' --raw |
Notes
--rawis mandatory — the human-formatted output is too wide to parse reliably.- Don't paginate; show the cheapest 10 and tell the user there are N more.
- Full query syntax lives in
skills/vastai/SKILL.md§ "Search".
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.
- 7d ago First seen · 43 lines · 14 tokens per session scan A 06bdd28a411f
search is a command published in the GitHub repository vast-ai/vast-claude-plugin (3 stars, last pushed 2mo ago), licensed MIT. It adds 14 tokens to every session and 537 once invoked, about $0.0001 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 2026-08-31.
Other commands, from other repositories
cost-optimize
You are a cloud cost optimization expert specializing in reducing infrastructure expenses while maintaining performance and reliability. Analyze cloud spending, identify savings opportunities, and implement cost-effective architectures across AWS, Azure, GCP, and OCI. Where provider-specific code appears below, adapt…
deploy
Deploy a frontend (React, Next.js, or static HTML) to a live URL on Butterbase.
integrate
Set up third-party service integrations.
env
Manage Vercel environment variables. Commands include list, pull, add, remove, and diff. Use to sync environment variables between Vercel and your local development environment.
scan
Scan AWS account for cost optimization.
finops-feedback
Step 5 (Feedback Loop & Celebration) — measure realized against projected savings, compute a labelled Cloud Entropy proxy, close the opportunity, and emit at least one new idea or policy update so the loop actually closes. Applies the double-loop gate. Mutates on the closure path.