In partnership with
Most coding agents answer like they're billing by the word. You ask where the bug is, you get a guided tour of your auth middleware, a history lesson on your dependency choices, and finally, buried in paragraph four, the line number. Today's lead fix is a skill that changes that habit: action first, steps numbered, no "hope this helps." It's a small file. It changes a real daily tax.
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.
|
Sponsored
|
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.
With Attio, you’ll get:
Leads automatically prioritised and routed to the right rep
Expansion and risk signals caught the moment they land
Follow-ups written in your voice, already there when you arrive
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
The Drops
Repoi-have-adhd
54,997 stars · ayghri/i-have-adhd
A skill to stop your coding agent from burying the answer. ADHD-friendly output, no diagnosis needed.
Here is the real problem it solves. Your agent explains itself for three paragraphs before it tells you which line to edit, and you reread the whole thing hunting for the fix.
Quick start: paste this into your CLI: Install the i-have-adhd skill/plugin from https://github.com/ayghri/i-have-adhd, refer to the repo's AGENTS.md for instructions.
Catch: it's a tone fix, not a reasoning fix. A bad answer stated plainly is still a bad answer.
AffiliateSaneBox
An agent that stops burying the answer only helps if the question reaches you, which is the same problem a cluttered inbox creates every morning. SaneBox, AI email triage that works on any inbox. It analyzes headers to decide what matters, says it never stores your message bodies, and moves low priority mail out of the way so what is left is signal. Fewer hours in the inbox, more hours building.
Use it for: cutting inbox hours on any IMAP, Exchange, or ActiveSync account, from a vendor that says it does not store your mail.
We may earn a commission.
RepoCodeGraph
73,397 stars · colbymchenry/codegraph
A pre-indexed code knowledge graph that auto-syncs on every change.
Use it for: giving Claude Code, Codex, Cursor, or Gemini surgical context instead of re-scanning your repo every turn, fewer tokens burned per question.
Catch: the hosted platform (test impact, break prediction) is a waitlist for early beta access. What ships today is the local graph, which is the part worth installing.
Repokeep-codex-fast
1,581 stars · vibeforge1111/keep-codex-fast
A backup-first skill that cleans up Codex's local state before it slows to a crawl.
Use it for: running in Inspect mode first (report-only, no writes) so you see what's bloated before you let it touch anything.
Catch: it archives, never deletes, which is the right default and also means you'll want to clear the archive yourself eventually.
MCPinvisible_playwright_mcp
2,644 stars · feder-cr/invisible_playwright_mcp
A Playwright MCP server that claims to be invisible to anti-bot and captcha detection.
Use it for: any browser-automation agent that keeps getting walled off by captchas before it finishes the task.
Catch: undetected automation is a sharp tool. Know the terms of service on anything you point it at.
|
From Our Partners
|
Take the prompts our creative team actually uses
The key to maintaining brand quality at scale? Mastering how to communicate with AI models and embedding them into your creative strategy.
Join our honest discussion with Ari Murray, Chief Digital Officer at Salt and Stone, and get 5 tips for expert LLM prompting with ready to run prompts.
Start Here
When your agent gets it wrong, tell it what's wrong instead of starting over.
Most people hit a bad answer and retype the whole question from scratch, hoping a fresh phrasing fixes it. It doesn't, because the model never learned what was wrong with attempt one, so attempt two just drifts into a new kind of wrong. Here's the better move.
1. Read the answer and find the specific part that's off, not the vibe, the exact line or claim.
2. Reply by naming that part directly: "the second step is wrong because X" or "this ignores that I already tried Y."
3. Do not restate your original question. The model has it.
4. Let it revise from there, one correction at a time, instead of one giant rewrite.
5. If it's still off, correct the new specific thing, not the whole response again.
TRY THIS: Open your last AI conversation where the answer missed. Instead of rephrasing your question, reply with just the one thing that was wrong about it, and watch how much faster the fix lands.
One term, plain English: LLM means a computer program trained on enormous amounts of text so it can predict and write language that sounds like a person wrote it.
|
Recommended
Once you are in the habit of naming exactly what is wrong instead of starting over, you want fewer, sharper inputs, and that goes for what lands in your inbox too. AI Chief, the newsletter on the AI tools worth your time rounds up the tools and breakthroughs that save you time so you stay sharp without reading everything yourself, free to subscribe. We may earn a commission. |
Frontier Signals
O'Reilly launched a tool that cites named authors instead of blending the whole internet into one confident voice. Its own 60-question benchmark found roughly one in five claims from leading AI tools unsupported by any cited source, and two-thirds of the sources that are cited name no author at all. (AIwire)
ChatGPT's new Intelligent UI answers with charts and buttons instead of a wall of text. It rolls out with GPT-6 to every ChatGPT tier, paid plans first and Free the next day, and you can dial the visuals back if you prefer text. (The Verge)
Three engineers let GPT-6 Astra drive a 2024 Toyota Corolla to an In-N-Out drive-thru, with a safety driver's foot over the brake. On their parking-lot benchmark it was the only model to finish the course; Claude got 45 percent of the way around and Grok 11. Nobody trained these models to drive. (Wired)
Attackers minted counterfeit TLS certificates for Google by compromising three country-code domain registries. Chrome now blocks every certificate Google has flagged, but the underlying weak link, who gets to issue a certificate, is still the same weak link. (Ars Technica)
Kubernetes co-creators Craig McLuckie and Joe Beda are building a cloud-native harness to keep coding agents reliable beyond the desktop. Claude Code famously began as a CLI tool, and McLuckie's pitch is that the value now moves from the desktop to the cloud. (Latent Space)
|
Recommended reading
If you like The AIgent, a small group of operator-tier publications worth your inbox: see the shortlist. |
Which best describes you right now?
Before You Go
Hit reply and tell me which Drop you're installing first, the i-have-adhd skill or the code graph. I read every one.
See you Friday.
Before you go: we started a room for people actually building with agents. Post what you shipped this week. Join the community →
|
Want to reach builders shipping with AI every weekday? Advertise with The AIgent. |
