11 JC Lab Ideas to Film, Test and Write Today: Browser Sandboxes, Screen-to-Agent Workflows and Two Crypto Attention Checks (2026)

OpenMAIC, Infragrid, and Sive.Host give you the clearest first filming blocks today: one turns a topic into a live classroom, one turns a screen recording into an agent runbook, and one gives you a measurable African hosting comparison. WebTerm Learn is the lowest-friction warm-up, while the two crypto entries are attention checks with a paper-trading boundary.

Video topics to film

OpenMAIC: turn one source document into a live AI classroom

Format: Tutorial

Film today: Clone OpenMAIC, install it with Node.js 20 or newer and pnpm 10 or newer, add one provider key, and run the local app. Upload a short public-domain article or paste a topic. Compare the generated slides, quiz, interactive scene, and spoken AI classroom. Finish by showing the editable PPTX or HTML export if the local setup exposes it. OpenMAIC describes itself as an open-source multi-agent classroom that can build slides, quizzes, simulations, and project-based lessons; the repository lists MIT licensing and the Node.js/pnpm prerequisites. 1

Hook: Can a creator turn one research link into a lesson viewers can actually interact with?

Boundary: A successful demo measures the local setup, provider, prompt, and source document. The demo gives you no production-quality test of every export path, classroom answer, or educational outcome.

Sive.Host: compare an African VPS with the European server you already use

Format: Comparison

Film today: Request the smallest comparable VPS you can access. Run the same ping, disk read/write, CPU hash, and cold-start checks from Lagos or your nearest location, then place the results beside Sive.Host's published baseline. The Hacker News post lists Johannesburg, 4 vCPUs, 8 GiB RAM, 200 GB SSD/NVMe, 4 TB bandwidth, 100 Mbps, and an IPv4 address at R103.08 per month, with a KVM Ubuntu 24.04.4 setup and its own ping and disk figures. 2

Hook: Does African hosting win on latency and value for an African creator, or only on the headline price?

Boundary: The published figures come from the provider's own post. Your comparison needs the same test script, region, plan size, and time of day; one VPS run cannot establish long-term uptime or support quality.

Infragrid: record a desktop task and inspect the agent's runbook

Format: Review

Film today: Record a repeatable creator task that crosses a browser and one desktop app, such as collecting three product links and placing them in a spreadsheet. Pause before any sensitive step. Inspect the generated runbook, change one approval rule, run it once, and show the action log and exception path. Infragrid says it follows work across browser and desktop apps, stores an encrypted Activity Timeline locally on the Mac by default, turns recordings into editable runbooks, and routes sensitive actions for approval. 3

Hook: Does a screen recording become a controllable workflow, or a replay that breaks as soon as the page changes?

Boundary: Use dummy accounts and synthetic data. Infragrid says the app needs Screen Recording and Accessibility permissions; the film should show those permissions and every external write before the run begins. 3

WebTerm Learn: learn one risky-looking command inside a disposable browser terminal

Format: Tutorial

Film today: Open the browser terminal, choose one Linux or Git lesson, and record the command before and after execution. Test one intentionally destructive-looking command only inside the provided environment, then complete a Git or Vim exercise and show the lesson's check. The official site presents WebTerm as a free online Linux terminal with interactive tutorials, no installation, and no sign-up. 4 Product Hunt's launch page describes WebTerm Learn as 12 courses and 129 lessons, all free, with a real browser sandbox that checks commands. 5

Hook: Can a beginner build terminal confidence without touching the files on their own computer?

Boundary: Treat the sandbox as a learning surface. A successful command proves what the hosted lesson allowed; it says nothing about the same command's effect on a local machine.

Tools and apps to review

Fotor Video Agent: edit the text after the AI has built the motion graphic

Format: Review

Film today: Give Fotor one short product brief with a price, two numbers, and a logo you own. Ask for a 20-second explainer, then change the number and headline on the multi-track timeline without regenerating the whole clip. Check whether the text, chart, and timing remain editable. The Product Hunt launch page says Video Agent turns ideas, scripts, or raw assets into motion graphics and keeps text, numbers, logos, and charts editable before rendering. 6 Fotor's official site lists AI video, image-to-video, product-video, and link-to-ad workflows, while leaving video-agent pricing undisclosed on the page retrieved today. 7

Hook: Can the tool save a creator from regenerating an entire video after one price changes?

Boundary: The editable-timeline description is a product claim. Test one short clip and record the actual export, account requirement, watermark, and render time before drawing a verdict.

Scientific Agent Skills: install a topical research subset instead of the whole library

Format: Tutorial

Film today: Pin version 2.65.0, install the repository with npx skills add K-Dense-AI/scientific-agent-skills, and enable one literature or database skill in the host you use. Ask the agent to find three sources on one narrow question, save the query trail, and compare one answer with the original paper. The repository lists 163 skills, more than 100 scientific and financial databases, MIT licensing, and compatibility with hosts including Cursor, Claude Code, Codex, and VS Code. 8

Hook: Does a curated research skill improve the source trail, or only make an agent sound more academic?

Boundary: The repository warns that skills can execute code, require review, and may need network access or API keys. Install one narrow subset and inspect its SKILL.md before granting access to a real project.

FreeLLMAPI: test the "free" routing claim with disposable keys

Format: Scam check

Film today: Run the local Docker setup in a disposable environment. Add no production credential. Send the same short prompt through two providers, inspect the X-Routed-Via header, watch the rate counters, and stop when the first quota or latency problem appears. FreeLLMAPI's README claims 34 free providers, 635 free model endpoints, and 7.4 billion tokens per month behind one OpenAI-compatible endpoint. The same README says the service has variable latency, no SLA, no frontier models, and a monthly catalog that can trail the live feed. 9

Hook: What does "free" mean after routing, daily caps, stale catalogs, and upstream providers enter the picture?

Boundary: The numbers are the repository author's claims, not an independent capacity audit. Use disposable keys, never paste private prompts, and report the observed provider, latency, and failure mode separately from the headline token count.

Packj: audit one dependency before you install it

Format: How-I-tested

Film today: Pick a disposable package manifest. Run the static audit first, then run the Docker --trace example against a package you are willing to isolate. Compare metadata warnings, static findings, and runtime behavior in three columns. Packj describes static analysis, dynamic tracing, metadata checks, and sandboxed installation across package ecosystems including NPM, PyPI, and RubyGems; the README gives Docker and GitHub Action paths and lists AGPL v3 licensing. 10

Hook: Does a dependency scanner catch behavior that a normal vulnerability list leaves out?

Boundary: Use a VM or disposable Docker environment. Packj marks Rust and PHP support as work in progress and says its default audit is static-only, so a clean static result is a screening result rather than a safety certificate.

Seeker / SKR: turn a 24-hour search spike into a liquidity checklist

Format: Trend react

Today's signal: CoinGecko's current 24-hour response includes Seeker, with market-cap rank 173 and trend score 1. CoinGecko defines this list as the top coins by popular searches and says the data refreshes every ten minutes. 11 CoinGecko's detail page identifies the asset as Seeker, ticker SKR. 12

Film today: Screen-record the trend result, then check the spread, 24-hour volume, order-book depth, and contract or exchange route available to you. Keep a paper entry and exit plan with a fixed loss limit. Recheck the same fields after 30 minutes and label the result as attention persistence or attention decay.

Boundary: Search popularity measures attention. It does not forecast price direction, liquidity quality, or a profitable trade. Keep the segment observational and use no real funds.

BOOK OF MEME / BOME: compare meme attention with tradeable market depth

Format: Trend react

Today's signal: CoinGecko's current response includes BOOK OF MEME, ticker BOME, with market-cap rank 378 and trend score 8. CoinGecko's detail page identifies the BOME/USD market page and the asset name. 11 13

Film today: Capture the trending rank, then compare the available volume, spread, slippage on a small paper order, and the number of accessible venues. Put BOME beside Seeker in one simple scorecard: attention, depth, spread, and exit friction.

Boundary: Meme-coin attention can move faster than a creator can record it. The scorecard describes a market snapshot and leaves investment decisions to the viewer; use paper trading only.

Article

Planetary Prediction Engine: what an AI geospatial workflow can do with Nigeria food-security data

Format: Article

Write today: Open with the practical question: what work remains when an AI agent can select geospatial data, engineer features, train several model families, and produce a report from a natural-language query? Use Google's August 27, 2026 research post as the main source. Google Research describes the Planetary Prediction Engine as an experimental system with three LLM-orchestrated stages: data selection, multimodal dataset curation, and automated model building. The post reports Nigeria food-security downscaling from state level to local-government-area level at R² 66.1% versus a 31.5% baseline, alongside results for U.S. public-health indicators and DRC Ebola nowcasting. 14

Angle: Explain the pipeline before discussing the scores. Then ask what a reader should verify: the benchmark definition, the baseline, the data available at prediction time, leakage controls, and whether an experimental research capability can be reproduced outside Google's data and infrastructure. The linked paper is available from the Google Research post for a second reading. 14

Boundary: The reported scores come from the authors' research post, which calls PPE early-stage and experimental. Write an evidence-led explainer, not a claim that the system is ready for operational humanitarian decisions.

Filming order

Start with WebTerm Learn for a fast browser capture. Move to OpenMAIC while the local environment is clean, then record Infragrid with dummy data. Run the Sive.Host comparison before the provider's baseline becomes harder to reproduce. Finish with Fotor, the two research-tool checks, Packj, and the paper-trading screen recordings. Write the PPE article after the benchmark table and its source trail are open beside you.