Agentic Coding Process for Planning Files

Opus 4.7 has been my daily model for 12 days. One workflow got clearly better. One got worse. I think both come from the same change. The better one: multi-step refactors land in a single pass. Last week I collapsed five specialized Claude Code skills into one shared rubric plus thin entry points — about 1,500 lines of overlapping rules across five files, consolidated into one canonical asset. With 4.6 that would've been a plan pass, then an edit pass, then a cleanup pass. With 4.7 it was one prompt and a clean diff. The worse one: trivial edits. Ask 4.7 to change a single line and it sometimes opens half the directory first, writes a planning preamble, and ships an edit with context it didn't need to load. The same instinct that made the refactor land in one shot makes the one-liner heavier than it should be. Same root cause: 4.7 plans further before acting. That's the upgrade and the regression at the same time. Which means scope discipline in the prompt matters more than it did three weeks ago. What I changed: - For one-line edits I now write "edit this line only, don't restructure" inline. With 4.6 that was usually unnecessary. - For multi-step work I dropped my old "first plan, then execute" scaffolding. 4.7 plans inside the model. My harness was double-counting it. The rule: the model's planning depth is a knob you have to actively dial down for trivial tasks. It's not a quality regression. It's a default that shifted under you. If you're using 4.7 for anything beyond a chat session, your prompts from the 4.6 era are calibrated to a different model. Re-check them. What's the first thing that broke for you on the 4.7 swap?

𝗧𝗵𝗲 𝗖𝗵𝗶𝗵𝘂𝗮𝗵𝘂𝗮 𝘀𝘆𝗻𝗱𝗿𝗼𝗺𝗲 (The art of creating stuff that's already there). - " Can you create a new reference for report xxx? " - " It already exists in the system, under the name zzz". All Document Controllers have had that conversation, and that is a part of our daily discipline. As a golden rule, you 𝗻𝗲𝘃𝗲𝗿 create before triple checking that it doesn't already exist. Duplicates are the ultimate enemy!💣 #DocumentControl

Opus 4.7 is no dud IMO. It's not only model step up, but also a harness update and needs to be treated differently - I've updated all of my processes and engaged differently with it and the results are so much better: Front-load prompts: Put intent, constraints, acceptance criteria, scope, and file paths in the first turn. Reduce turn count: Batch questions instead of drip-feeding instructions mid-task. Use effort levels intentionally: Default to xhigh for coding/agentic work, lower for mechanical tasks, raise to max only for genuinely hard problems. Migrate to adaptive thinking: Stop using fixed thinking budgets; steer depth with prompt wording instead. Rewrite prompts for literalness: Make scope explicit: what applies everywhere, what is out of scope, what not to touch. Separate finding from filtering: In code review, find everything first, rank/filter after. Be explicit about tools/subagents: Tell it when to parallelize, when to read files, when not to delegate. Manage context more actively: Compact earlier, clear stale sessions, use subagents when only conclusions matter. Trim ritual/overhead prompts: Remove verbose checklists and multi-step clarification habits for small tasks. Drop deprecated patterns: Stop relying on assistant prefills; recheck any old harness assumptions.

I built a Transcriber plugin for @obsdmd ✍️ Point it at any image in your vault and get structured Markdown back. Headings, lists, quotes, even Mermaid diagrams. Runs 100% locally via Ollama. No data leaves your machine. Step 2 of my handwriting-to-text pipeline. Watch: https://lnkd.in/eiq_C5jA

100 active users. 100 engineers who just fired Excel for their Rate Analysis. 🚀 When we initially coded QuickBOQ Pro for our internal projects at NSA Constructions, the goal was entirely selfish: We wanted to stop wasting hours manually typing CPWD coefficients and hunting through 400-page DSR PDFs. We did not expect 100 other contractors, consultants, and billing engineers to jump on board this quickly. Over the last few weeks, this platform has helped users generate automated, error-free estimates, cutting their BOQ preparation time from hours to literally minutes. Why are engineers switching? Because the industry is exhausted by: ❌ Broken Excel formulas. ❌ Getting 'Extra Item' rates rejected over manual errors. ❌ Guessing market rate impacts during volatile price shifts. QuickBOQ simply auto-fetches the standard coefficients, calculates the 1% Water Charges and 15% CP&OH, and exports a perfect "Justification of Rates" PDF ready for departmental submission. A massive thank you to the early adopters who tested the system and pushed for the latest DSR 2025 database integration. Your feedback built this. If you are still calculating your tender rates manually, come join the 100 professionals who have already automated their workflow. Try it here (Link in comment) #CivilEngineering #QuickBOQ #ConstructionTech #Estimation #CPWD #FoundersJourney

No More F-File Names in A-Files. 🚫📂 ------------------------------------------- A-Files = Administrative Files In any system, clarity is power. Administrative files are meant to be structured, searchable, and efficient. But when “F-type” naming creeps in — random, unclear, inconsistent — it creates confusion, delays, and mistakes. 💡 What to avoid in A-Files: ❌ Fuzzy names (no clear meaning) ❌ Fragmented formats (no naming standard) ❌ Forgotten context (missing dates, versions, owners) 💡 What to implement instead: ✅ Clear, consistent naming conventions ✅ Standard formats (Date_Project_Purpose_Version) ✅ Easy traceability and quick retrieval Because in administration, seconds lost in searching = productivity lost at scale. Discipline in file naming isn’t small — it’s the foundation of organized thinking and execution. 📌 Fix the names. Fix the system. ------------------------------------------- #divine #kalammalai #motivation #inspiration #technology #entrepreneur

A common question we’ve been receiving lately: “What happens when a user visits a website, finds an inaccessible PDF, and wants an accessible version?” This is becoming increasingly important as accessibility awareness grows. Here’s how an effective workflow can look: • The user requests an accessible PDF through a form/button on the website • The request is routed to the service provider • The PDF is shared with a remediation team • The document is remediated as per accessibility standards (WCAG/PDF/UA) • QA is completed • The accessible version is delivered back within a defined SLA At StarAble Docs, we help organizations implement this end-to-end process smoothly. If you're planning to add an accessibility request mechanism to your website, feel free to connect with us. 📩 faiza@starabledocs.com 🌐 https://starabledocs.com

By now, we’ve all seen the Anthropic leak. If you’re trying to make sense of it, this is one of the clearest breakdowns I’ve come across. It maps how Claude Code works under the hood, from the agent loop to its tool system, and how tasks are orchestrated. Super helpful if you’ve been trying to understand how these coding agents actually work in practice. Worth a look 👇 https://ccunpacked.dev/

If you’re not reviewing sign-in sheets, you’re flying blind on labor loading. Most projects treat sign-in sheets like compliance paperwork. The value is in the weekly review. If the schedule assumes 15 workers and you’re consistently seeing 6, you don’t have a “productivity” problem. You have a planning + accountability problem. And the sooner you're aware of it, the more time you'll have to collaborate with the team on fixing it. For developers, sign-in sheets also help with: - lender/insurance documentation (who was on site) - dispute defense (who was there when an issue happened) - sanity checking the GC’s narrative against reality

Creating Agentic Skills with CLI feels so much better than building MCPs! I have built and SRT file parser to reduce context usage when information is the only thing that matters. No need to feed notes, when the exact story can be shared. It unlocks: feedback, assistance and improvements.