Automate PDF Generation with Bookmarks

☕️ Introducing PQL, Plane Query Language, and enhanced Dashboards. Now generally available. Plane's standard filter UI handles most cases. PQL is for when teams need compound conditions, nested logic, or queries that mix custom properties with built-in fields. - The syntax is Field Operator Value, with autocomplete for fields, operators, and values. Save any query as a named view. For anyone who'd rather skip the syntax, Text to PQL turns a plain-English description into a query you can run or edit. - PQL extends into dashboards at three levels: source filters, quick filters, and widget filters. Same syntax across all three. - Also shipping: redesigned dashboards, a table widget, and interactive charts you can drill into. It is day 2 of our caffeinated launch week. ☕️

If you use Claude Code daily and only know 5 of these 32, you're leaving most of its value on the table. 9 slides. Reframed for those using Claude Code for their business rather than those shipping code on it. Some examples from each slide: 2: the safety net stack that stops you watching Claude rebuild the same thing three times 3: CLAUDE.md routing - the cheapest leverage in your entire system prompt 4: the Haiku/Sonnet rule that cuts cost without tanking quality 5: the command everyone still types to make Claude think harder, and what actually does it now 6: the one MCP install that defeats Claude's training cut off in real time 7: the worktrees syntax that's been mis-typed everywhere for months 8: the safer autopilot mode shipped in March that most haven't switched to yet 9: power moves - custom skills, plain English data analytics, and the comment magnet If you use Claude Code for your business, the 27 you don't know cost you 10x more than the 5 you do.

The Wow effect of building tools (MCP, plugin) for coding agents? Multimodal input. I am working with a Motley customer to set up a complex reporting automation. Their setup for our test: - 3 tables with 370 columns and 640,000 rows - 3 models of 300+ dimensions and measures - A report with 30+ charts and tables to generate A couple of graphs were off, mostly because I did not really understand how they were constructed. Here is how I prompted Claude to update the queries of these charts: "Go fix it using the Motley MCP. Here is the context" - the transcript of a call with our customer on what two of the graphs represent and where the data is - Yaml configs for two other graphs - screenshots of dashboards showing the reference result. 💥 : one shotted and with a report now ready to be generated in Gamma. Got to love these crazy thinking machines.

One file. 300 pages. 150 customer records. What happens next? Usually: → manual splitting → saving files → renaming 'Word – Split by Text' automates that entire step. Define your marker → get individual documents → continue the flow. This is where automation actually saves time. https://lnkd.in/eYDhHBN4

Open ~/.claude/ on your machine. Count the skills, slash commands, hooks. Do you actually know what loads when? Neither did I, for a while. My setup grew to plenty of skills and a dozen slash commands before I admitted I had no idea what was in context. That’s where the idea is born, that a vector store will fix it, aaand… WRONG 👊 it won’t, because the problem is scope, not retrieval. Scope is the thing to manage. What loads when, who writes it, how long it lives. Five pieces to start with: 👉 MEMORY.md is the routing index. Under 150 lines, auto-loaded every session. Everything else gets pulled on demand. 👉 ACTIVE.md holds persistent operational state - WIP, blockers, recent decisions. 👉 SCRATCHPAD.md is mid-session working notes. 👉 wiki/ holds curated knowledge by domain 👉 raw/ collects auto-captured insights. At session end a Stop hook reviews scratchpad - operational stuff goes to active.md, insights go to raw/, rest gets wiped. The raw/ → wiki/ pipeline is Karpathy’s living knowledge base. The five piece split by lifecycle is an extension. Each piece decays at a different rate: • MEMORY.md is stable • active.md churns • scratchpad.md dies with the session • wiki/ grows slowly • raw/ fills up every session and gets drained Been shipping with this for months 🚢 No skills directory, no plugin bloat, no slash commands beyond occasional /memory-compound. Just five files and a Stop hook. Works better than every “10k stars in a week” memory system I’ve tried so far. DM me if you want the ready full prompt to autogenerate it. Or just rebuild it from the structure.

Tired of reading large nested API responses? 🤯 What if you could turn raw response data into searchable, filterable UI components instead of scrolling through endless JSON? Response Visualizer in Bruno helps you transform API responses into beautiful tables and custom HTML dashboards — making it easier to explore, filter, and analyze data during API testing. ✨ Build searchable tables ✨ Create custom dashboards ✨ Visualize response data instantly ✨ Improve debugging and testing workflows Stop reading raw JSON. Start visualizing your API responses 🚀 Read Blog post here: https://lnkd.in/gzmUypcj

atlon.io supports #FOCUS 1.3 validation for both Cost&Usage and Contract Commitment datasets! What changed since 1.2: The headline addition is a second dataset: ContractCommitment with 13 mandatory columns describing the contracts and commitments your charges roll up to. CostAndUsage itself grows from 57 to 65 columns: HostProviderName and ServiceProviderName join the mandatory set, and six new conditional columns cover split-cost allocation plus contract linkage (ContractApplied). How to validate? Head to https://atlon.io/, drop a CSV or Parquet file and pick your FOCUS version. For 1.3 you also pick the dataset (CostAndUsage or ContractCommitment). Choose Schema mode for an instant column-presence check, or Full mode to run every rule (type, format, nullability, business logic, the lot).  

In this week's video, we look at four practical Power Query tips that change how calculations behave. By making small tweaks to the generated M code, we can produce very different results and create simple solutions that would have been impossible with the user interface alone. Based on the comments I've received so far, there are some real nuggets of gold in there. 📺 Check out the video here: https://lnkd.in/eSf3cdya

Started with the simplest possible agent: one LLM call in, one text response out. Then I layered in tools, one capability at a time: web_search + fetch_page — research and source gathering save_answer — cited Markdown generation, persisted to disk render_image — image generation via Higgsfield pin_to_dashboard — publishing cards to a live Prefab UI Now a single prompt — "Find 10 stocks worth buying, generate an image, and pin it to the dashboard" — fans out into the full chain automatically. The agent decides which tools to call, in what order, and how to thread the outputs through. The same tool layer is exposed as a standalone FastMCP server, so it's reusable across agents, interfaces, and workflows — not locked to a single host. And to make the whole thing legible, I built a live dashboard that visualizes the agent's reasoning in real time: which tools it picks, how it chains them, and how the final output is assembled. One prompt in. Research, generation, and publishing out. End to end.

𝗗𝗮𝘆 𝟯/𝟮𝟬 — 𝗔𝗜 𝗣𝗿𝗼𝗱𝘂𝗰𝘁 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 𝗖𝘂𝘀𝘁𝗼𝗺 𝗥𝗔𝗚: Built an automation workflow for a custom RAG system: • Captures user inputs — details, files, and preferences • Supports multiple document types such as PPT, CSV, and HTML • Performs effective chunking for each use case and stores embeddings in the Pinecone vector database • Performs use-case-based retrieval involving separate user prompts and retrieval techniques 𝗞𝗲𝘆 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴𝘀: • LLM routing is essential, as agent behavior differs for each use case • Understanding the working of RAG and Agentic RAG • Design decisions required to improve performance 𝗣𝗲𝗻𝗱𝗶𝗻𝗴 𝗜𝗺𝗽𝗿𝗼𝘃𝗲𝗺𝗲𝗻𝘁𝘀: • The latency for summarization and questionnaire generation can be improved Automation Link:- https://lnkd.in/dCUV2mNq