Activity
Mon
Wed
Fri
Sun
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
What is this?
Less
More
Clief Notes
46.2k members • Free
127 contributions to Clief Notes
27d •
Ronin Loop
My agents were lying to me and I couldn't tell Retired NYPD detective. I run three businesses on Codex/Claude Code agents across three Macs. Last week I found out my monitoring layer had been reporting healthy while doing nothing, for months. Not broken. Lying. Different problem, and it's the one nobody talks about. Here's what a day of tearing it apart taught me, and the loop I built out of it. The failure nobody names Nine distinct ways my system reported success while failing. A few: - A wrapped job that exits 0 having done nothing gets reported GREEN. ~135 jobs affected. - A missing receipt is indistinguishable from a successful run. A job that never ran looks identical to one that passed. - Drop any `.json` file into a receipts directory and you fabricate a clean bill of health for 39 states. Worse — a stray file with a newer timestamp *displaces* a real failure record. I had a genuine timeout that got overwritten and turned green. - My alerting worked perfectly. It delivered six RED alerts over six days. Nobody read them. That last one matters most. The system wasn't blind. It was ignored. Alert fatigue is a design failure, not a discipline failure. Meanwhile a detector I'd built to catch exactly this had been running for ten days, correctly identifying that no leads had shipped to my ops manager — and telling nobody, because delivery was never armed. A working alarm wired to nothing. --- The loop Every agentic framework I read describes the same thing: an agent that generates, checks, and repeats. What none of them specify is what happens when the check says no. That's where mine lives. ``` ROUND N: 1. CONFIRM reproduce the defect behaviorally, never fix on description 2. FIX smallest diff at root cause, harden against the class 3. SELF-TEST builder proves it. Verdict here is PROVISIONAL. Always. 4. REGRESSION re-run every prior test, PLUS generate adversarial inputs against the surface this round just changed 5. AUDIT hand to a different model as an OPEN QUESTION —
0 likes • 6h
@Jim Galvin you wrote that one of your panels "logged a specific failure mode to its own reference doc, then committed that exact mistake an hour later". We've hit that three times in a Claude Code setup on Windows, in a completely different kind of task, so I think what you've got there travels further than one stack. Ours was a drafting quality rule, nothing to do with infrastructure. The third one was worse than yours. The agent read the reference file specifically to avoid the failure mode, declared the draft clear, and then shipped the exact thing the file names. Having read the file it was more confident and no more accurate. What has actually caught it for us isn't the warning, it's a forced output. Before declaring anything clear, quote the lines that could be violations and say why each one passes, and treat zero named lines as the signal to look again. Silence becomes a FAIL rather than a PASS. Which is your exit condition, I think, just pointed at governance instead of infrastructure. A warning sitting in a reference doc is trusted because it's present, the same way your receipts were, and the same way any executable was a reviewer. One other thing, on step 5. You describe handing it to a different model. In our own testing the model swap gave us close to nothing in the way of independence, both reads came back the same anyway. What changed the answer was the framing you've already got in there, 'present or absent' rather than 'confirm this'. So the independence is coming from the question. You could probably drop the second model and keep the framing, then see whether the answers hold. Cheaper if they do. And the rule you hold hardest, that nobody validates their own item, is an evidence rule. Reads like the detective in you rather than the engineer. Happy to go into the specifics of the forced output check if that's useful to you.
4d •
From pdf extraction to customer signatur
Hi everyone! Finally, I feel like I can share what I’ve been building behind the scenes. AND I AM FUCKING PROUD OF THIS ONE! 🔥 What do you think something like this is worth? I’ve built an end-to-end contract automation pipeline that runs 100% inside the Microsoft ecosystem. The best part? There is ZERO AI in this flow. Just pure, deterministic, enterprise-grade automation that works flawlessly every single time—and costs next to nothing to run! ⚙️ How the Flow Works: Drop & Trigger: You drop a .pdf file into a specific SharePoint folder. This instantly triggers a Power Automate flow that extracts the data and generates a draft contract. Interactive Teams Approval: The person who uploaded the file receives an automated notification in Microsoft Teams. Right inside Teams, they can: - View the uploaded source file. - Review the generated draft contract. - Approve or deny the contract. - Edit key values directly (edits apply to the final executed contract). E-Signing (No Third-Party Fees): Once approved, the contract is automatically emailed to the customer with a unique link to review and e-sign. - Customers can sign using a mouse, touch screen, stylus, or by uploading a signature image. - Note: You can plug in Adobe Sign or DocuSign, but why pay per signature during testing when you can build it yourself? 😉 Legally Sound Audit Trail: The system captures and logs full verification parameters: [Signed By], [Signed At], [Signer Email], [Signer IP], [Signer Device], [Signature Method], [Document Hash], [Reference], [Approved By], [Approved At]. Final Delivery & Auto-Cleanup: Once signed, the finalized contract is emailed to both parties. The flow then automatically deletes all temporary input and output files, leaving a clean footprint where only the two emailed contracts exist. 🛡️ Built for Enterprise & Clients Bulletproof Reliability: Built-in safety features prevent duplicate entries, invalid file formats, or missing data fields.
2 likes • 3d
@Allan Durhuus, super clean design. Zero tokens burnt on each round, and you know exactly what you get every time you press the button. Consistency is critical in B2B. On what it’s worth. Ballpark, and priced to get into the market rather than at the ceiling. First bespoke deployment into a client tenant, €8-12k. Redeploys off the template after that, €3-5k, because you’re mapping paths rather than building. Small monthly on top for support. Against DocuSign seats and the admin hours it takes out, it pays back inside a year. One worth a look while you’re in there. You’re logging signer IP and device. In the EU that’s personal data, so it sits under GDPR separately from the signature question. You’re already auto-deleting the temp files, so a retention rule on the audit log is the same move applied just one layer down. On @David Vogel’s point, the consent disclosures are worth having, but they’re ESIGN and that’s US law. You’re in the EU, so you’re under eIDAS, which works differently. No paper copy right, no withdrawal notice. Article 25 says a signature can’t be denied legal effect just because it’s electronic. Where David’s point does land is at deployment. If you’re exporting this into a US client’s tenant, that’s when the disclosures get mapped in, same as the SharePoint paths and API keys. Template stays as it is.
9d •
🤯 Something in here has been broken for a while
Alright, here’s what we’ve been sitting on. We’re past 2,700 posts and 40+ lessons now. The number one thing we hear on Wednesday onboarding calls is some version of: “Where do I start?” “What should I do next?” “Did someone already answer this?” Fair. There’s a lot in here. So we’re bringing in an AI trained on the Clief Notes library. Every lesson. Every Afternoon Tea. Every High Tea drop. It actually knows what’s in here. 🧭 Tell it what you’re stuck on, and it points you toward the right lesson or drop. 🗺️ Tell it where you’re at, and it can help you figure out what to focus on instead of just telling everyone to “start at The Foundation.” 🔍 Ask it about something we’ve covered before, and it helps you find it without scrolling through months of content. And when we launch it, everyone gets access. More on that very soon.
1 like • 8d
@Alex Brown would be a great 'Show Your Work' contender...?
9d •
The only training video for my company
Remember sound!!!! It has been a long time since I posted last and a lot is happing right now! But I wanted to share the only training video I need to show everyone in the company on how to use my automations.
1 like • 9d
@Allan Durhuus good to hear from you and sounds like things are going well. Are you going to be producing a manual to go along with the training video? All jokes aside very interested to hear more as I work with SharePoint and have done since WSS days. Any show your work post plans…..?
18d •
They work WITH the ICM
Feed these papers to your ai and ask it if it thinks they work well WITH the ICM. They aren't here to replace anything. They are the gang ICM didn't know it started. OGM: https://doi.org/10.6084/m9.figshare.33106949 EEM: https://doi.org/10.6084/m9.figshare.33106970 OACM: https://doi.org/10.6084/m9.figshare.33106985 OGM + EEM Bundle: doi.org/10.6084/m9.figshare.33107024 OACM Bundle: doi.org/10.6084/m9.figshare.33107030
0 likes • 10d
@Andre Cordero went back and read your closure gate post from a couple of weeks back This is the bit I wanted to ask about. "The problem was never that I can't judge 'done', it's that I was re-judging it manually every time, and the agent's 'all tests passing' was wearing me down into rubber-stamping." That matches something I've been watching in my own work, and I hadn't seen anyone else write it down. You've already headed off the obvious objection, so I'm not asking that one. Your own note says OACM never claims accepted work is correct, only that acceptance followed the frozen contract and admitted evidence. Fair comment. What I'm less clear on is what tells you the contract has gone stale. The agent's "all tests passing" was a conformance claim against its own idea of done. CONTRACT.md moves that up a level, and the gate's verdict is a conformance claim against a document. In both cases the thing nobody re-reads is the standard itself. And if you wrote the contract on the fortieth verification of the day, the tiredness is in the file, and now it gets applied every run instead of once. Separate question. You've asked people to break your work a few times now, on the gate, on the recurrence repo, and again to me on this thread. Has anyone actually done it? From outside it looks like the offer keeps getting taken up as a design conversation rather than an attempt to break anything, and I wondered whether that matches your experience. On your offer, I've got it noted and I'll come back to you on it.
0 likes • 9d
@Andre Cordero yes, go ahead with the acknowledgment and the issue. Thanks for asking rather than just doing it. Also, "thinking about it a bit changed my mind about the fix" isn't a common thing to say out loud. Fair play to you. I'd rather not hand you a tripwire though. It's your loop and you'll design it better than I will from out here. One thing I'd want to be clear on before I'd trust one. Your OACM section 4.10 already handles support going bad after you've accepted it, right down to a defeater reopening when the evidence that resolved it dies. So is a stale contract a different problem from that, or the same problem pointed at the contract instead of at the evidence? I genuinely don't know which. Your break it offer is still noted, by the way. Not forgotten.
1-10 of 127
@andrew-carter-8893
Ideas to execution with AI 'factories'. Turning theory into practice with build, iterate, refine, learn.
Online now
Joined Apr 20, 2026
Powered by