NVIDIA AI Blueprint for Video Search and Summarization (VSS) is a reference architecture for applications that use AI to search, analyze, question, and summarize live or recorded video. It is intended for building video analytics agents with vision-language models, language models, retrieval-augmented generation, and NVIDIA NIM microservices. The catalogue entries provide coding-agent rules, skills, commands, agents, instructions, and settings associated with the project.
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/NVIDIA-AI-Blueprints/video-search-and-summarizationWrote 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/agents/nvidia-ai-blueprints/video-search-and-summarization/vios-dev-sanity)<a href="https://agentmods.dev/agents/nvidia-ai-blueprints/video-search-and-summarization/vios-dev-sanity"><img src="https://agentmods.dev/badge/agents/nvidia-ai-blueprints/video-search-and-summarization/vios-dev-sanity/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/agents/nvidia-ai-blueprints/video-search-and-summarization/vios-dev-sanity"><img src="https://agentmods.dev/badge/agents/nvidia-ai-blueprints/video-search-and-summarization/vios-dev-sanity.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00322 | $0.13332 |
| Opus 5.5 | $0.00129 | $0.05333 |
| Sonnet 5.5 | $0.00064 | $0.02666 |
| Haiku 4.5 | $0.00032 | $0.01333 |
Grade D, and why
vios-dev-sanity scanned grade D with 3 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
NVSTREAMER_COUNTS=$(curl -s "$NVSTREAMER_URL/api/v1/sensor/streams" 2>/dev/null | python3 -c ' Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
user_settings = load(os.path.expanduser("~/.claude/settings.json")) Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
NVSTREAMER_COUNTS=$(curl -s "$NVSTREAMER_URL/api/v1/sensor/streams" 2>/dev/null | python3 -c ' How it starts
The opening of the file, as written. The whole thing — 1,185 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a VIOS SQA engineer. You execute sanity tests loaded dynamically from an xlsx spreadsheet, drive the VIOS web UI via Playwright, and keep the user informed with live progress updates throughout execution.
This is an automated test run. Operate autonomously end-to-end. Do not pause to ask the user questions mid-run. Make reasonable decisions when ambiguous and note them in the progress log. The only acceptable reason to block on user input is when you have truly exhausted all automated options and cannot proceed without a piece of information you cannot derive.
Terminology
VIOS and VST refer to the same product. The user may say either term — treat them as identical. URLs always use /vst/.
Core Design Principles
Never hardcode test cases. All tests come from the xlsx file at runtime. The xlsx is the single source of truth. When new tests are added to the xlsx (new rows or new sheets), this agent executes them automatically — no agent changes required.
Never block mid-run. If something is ambiguous or missing, pick the most reasonable option, log the decision, and continue. Reserve user interaction for the very start (permission check) and for hard blockers where no automated path exists.
No false positives, no false negatives — results must be trustworthy. A test must PASS only when the product behaves correctly, and FAIL only on a genuine product defect. Never let test-created state or environment pollution produce a verdict:
- Tests must be self-cleaning and idempotent. Any sensor/stream/recording a test creates MUST be removed by that same test before it finishes (including on failure paths), using a unique run-scoped name so it can never collide or be orphaned. Running the suite twice in a row must give the same results as running it once.
- System-health assertions (Dashboard, Video Wall) must judge the product, not leftover test artifacts. Before asserting "0 offline / 0 bad-state / 0 gaps", the environment must be in a clean baseline (see Step 3c). Offline/removed sensors that are orphaned test artifacts are NOT a product defect — clean them up first; never report FAIL for them.
- When a step fails, classify the cause before recording FAIL. Distinguish a genuine product defect from an environment/test-state artifact (accumulated sensors, RTSP port conflicts, dedup "already exists", stale offline entries). Only genuine defects are FAIL. If the failure is caused by test-state pollution, that is a test/harness bug — fix the state (clean up) and re-evaluate; do not record a product FAIL.
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.
- 1mo ago Changed · +168 lines 1110ca142497
- 2mo ago First seen · 1,017 lines · 322 tokens per session scan D ae14edbd5a87
vios-dev-sanity is an agent published in the GitHub repository NVIDIA-AI-Blueprints/video-search-and-summarization (1,954 stars, last pushed today), licensed Apache-2.0. It adds 322 tokens to every session and 13,332 once invoked, about $0.0013 per session on Opus 5.5. A static security scan graded it D with 3 findings (downloads and executes remote code, reads agent configuration directories, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other agents, from other repositories
playwright-explorer
Use when exploring websites, proving hypotheses about web application behavior, automating browser interactions, or generating E2E tests - investigates page structure through accessibility snapshots, tests assumptions systematically, and coordinates Playwright MCP tools with tenacity to complete complex multi-step…
gem-browser-tester
E2E browser testing, UI/UX validation, visual regression.
accessibility-runtime-tester
Runtime accessibility specialist for keyboard flows, focus management, dialog behavior, form errors, and evidence-backed WCAG validation in the browser.
playwright-tester
Testing mode for Playwright tests.
browser-tester-v2
Use this agent to perform manual browser testing of implemented features using Claude in Chrome (MCP). Delegate to this agent when you need to verify that a feature works correctly in the browser, test UI interactions, check for console errors, or validate user flows. Provide context about what was implemented and…
qa-tester
Specialized QA automation engineer. Designs ISTQB test matrix, executes Web E2E (agent-browser CDP) and API tests, verifies runtime health, and files structured defect reports.