🛠️ TradingAgents Won't Let a Backtest Cheat

In partnership with

TradingAgents just crossed 110,000 stars, and the part worth your attention is not the number. It is that the last five releases all fixed the same sin: letting a backtest see data from the future. Point-in-time integrity, SEC filings served as they were actually filed, past decisions settled while the analysts work. Plenty of multi-agent demos optimise for confidence. This one spent five versions removing ways to cheat.

One reader pays five dollars a month so the daily stays free for everyone else. If it has earned a spot in your morning, be number 2.

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.

LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.

The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.

Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

The Drops

RepoTradingAgents

110,269 stars · TauricResearch/TradingAgents

A multi-agent LLM framework that runs trading decisions through separate analyst, researcher, and manager roles before a trade ever executes.

Here is the real problem it solves. A backtested trading bot can quietly cheat by letting later data leak into earlier decisions. This one enforces point-in-time integrity across SEC filings, macro data, and social sentiment, so a backtest only sees what existed on that date.

Quick start: clone TradingAgents.

Catch: a provider per model tier means you are choosing and paying for multiple models just to run one decision cycle.

Open the repo

AffiliateHubSpot

TradingAgents earns its stars by refusing to let an agent see data it should not have. Your customer data needs the same discipline: one record, dated, that an agent can act on. HubSpot free CRM, a contact database with email, deal pipeline and forms on one record, free with unlimited users. If your customer list is a spreadsheet and your follow-ups live in your inbox, this is the step before any agent can act on them. The data has to sit somewhere first.

Use it for: moving the customer list out of a spreadsheet onto one record an agent can act on.

Start with the free CRM

We may earn a commission.

RepoAutoHarness

9,591 stars · tigerless-labs/autoharness

A self-learning skill layer for Claude Code that distills skills from your real sessions instead of a separate training loop.

Use it for: letting your agent build its own playbook as it works, then merging near-duplicate skills instead of piling them up.

Catch: it only touches the skills it wrote itself, so it will not clean up whatever mess your harness already has.

RepoAnswer me with HTML

2,319 stars · QingYunA/answer-me-with-html

An agent skill that turns a hard question into a one-page HTML answer instead of a wall of markdown text.

Use it for: asking your agent to map a codebase or explain a protocol and getting something you can actually read about 50 milliseconds later, instead of a page that cost 4,900 tokens to hand-write.

Catch: the gain only shows up on explainer-style answers. Simple factual questions do not need the extra render step.

ToolOpenEdit

1,602 stars · veedstudio/open-edit

An open-source, agent-driven video editing pipeline with no GUI and no timeline. You describe the edit, your coding agent renders the file.

Use it for: adding subtitles, motion graphics, or slides to a video clip by typing a sentence into Claude Code, Codex, or Gemini CLI.

Catch: hosted transcription needs a VEED account, and the free tier covers about ten minutes of audio a month.

From Our Partners

Stop typing what you could say in 10 seconds.

Wispr Flow turns your voice into clean, professional text inside any app. Emails, Slack, client updates: speak once, send without editing. 4x faster than typing. Works on Mac, Windows, iPhone, and Android.

Try Wispr Flow free

Start Here

The meeting recap you stop writing twice.

One prompt today instead of a how-to. It is one of ours, tested on ChatGPT and Claude with full notes and with details missing. Copy it, paste it into the free app you already use, and hand it the notes from your last meeting exactly as you typed them.

You are the person who writes the recap email after every meeting, and you never invent what the room did not settle.

Before you write, if any field below is empty, ask me for it and wait.

Here are my notes from a meeting that finished a few minutes ago. They are unedited and some lines are half sentences.
[MY NOTES, PASTED EXACTLY AS I TYPED THEM]

Who was in the meeting: [FIRST NAMES]
Who else is getting this email, and what they need from it: [WHO, AND WHAT THEY NEED]
What was already settled before today: [ANYTHING NOT TO REOPEN, OR "nothing"]
Today's date: [DATE]
How I sign off: [SIGN OFF]

Turn this into one email that works for both lists, so I do not have to write it twice.

Give me, in this order:
1. A subject line that names the meeting and the one thing that changed.
2. Two sentences for someone who was not there: what the meeting was for, and what came out of it.
3. What was decided. One line each, and only things my notes show were settled.
4. Who is doing what by when. Use the names I gave you. Where my notes name no owner, write owner not named. Where there is no date, write no date agreed.
5. What is still open, written as questions.
6. One closing line telling people what to do if something here is wrong.

Rules:
- Do not invent a decision, an owner or a date that is not in my notes.
- Do not turn a discussion into a decision, and do not soften a decision into a discussion.
- Something settled before today, or the customer's ask, is context for section 2, never a line under What was decided.
- Someone who missed the meeting should understand it without asking me anything. Someone who was there should find nothing new in it.
- Under 300 words, plain sentences. Sign it with my sign off.
- If my notes show no decision at all, say so and write the email as a summary of where the discussion got to.

Before the email, give me one line naming anything in my notes you left out, and why.

The line that makes it work: name anything you left out, and why. A model does not summarise by default. It folds everything in and hands your notes back with headings on. Making it say what it dropped shows you whether it chose or just reformatted, and that is the habit that keeps every AI answer honest.

TRY THIS: run it on your last meeting, then read the dropped line before you read the email.

This is one of 100 in the AIgent prompt pack. The other 99, plus guided builds with Mike on video, come with the AIgent membership, opening next week.

Want in when the AIgent membership opens next week?

Recommended

Once the recap is written, the owners and dates in it need a home that is not another email thread.

ClickUp, one workspace for tasks, docs, and chat brings your tasks, docs, chat, and goals into one place so a small team stops switching between seven apps, free to start.

Try ClickUp free →

We may earn a commission.

Frontier Signals

OpenAI published results on 377 open math problems, produced by an internal model it has not released. Operators building reasoning-heavy agents should watch which techniques get published, not the problem count. (Fortune)

A suspected China-based attacker used ARTEX, an open-source agentic pentesting tool, with Claude on the side, in a campaign that pulled data out of South Korean banks. If your stack runs an autonomous tool-call loop against production systems, your abuse-monitoring needs to assume the attacker has the same agent you do. (Infosecurity Magazine)

Manus raised more than $500 million in its first round since regulators forced Meta to abandon the acquisition. The agent-startup market held up even after a high-profile deal collapsed, which tells you funding sentiment for agent tooling is sturdier than one failed merger. (CNBC)

US venture deal value hit a record $515.8 billion in the first nine months of the year, and AI accounts for 82.7% of it. Concentration this heavy means a slowdown in frontier-lab funding would ripple through every tool built on top of it. (SiliconANGLE)

USA Today and the local papers it owns sued OpenAI for copyright infringement, seeking more than $250 million in damages. Add it to the pile of publisher suits. If you are building on model outputs trained on scraped news, the legal exposure sits with the lab, but the product risk sits with you. (The Verge)

What are you building with right now?

Before You Go

Reply and tell me which one you are cloning this weekend, TradingAgents or AutoHarness. I read every reply.

The point-in-time fix in TradingAgents is the whole lesson this week: an agent is only as honest as the data you let it see. Audit what your own pipeline leaks before you trust what it tells you.

See you Monday.

Before you go: we started a room for people actually building with agents. Drop your best Claude Code skill. Join the community →

Want to reach builders shipping with AI every weekday? Advertise with The AIgent.