Kolbe Yang

@plannotator @linear Can review agents leave comments or respond to my comments in the UI? Something like hunk.dev Similar but a bit I think different to Guided Reviews.

@linear Pi and Copilot join the provider ensemble (for guided reviews and general agent reviews). Reminder: you can launch your own custom review skills from the UI.

Plannotator 0.22.0 is out. Code Review (is not dead yet): - Guided reviews give you a list of changes with semantic overviews, ordered by importance & where you need to spend your time reviewing. Inspired by @linear, this works for local and PR changes . You don't need the best models or high reasoning levels to use this! - Pi and Copilot are added as agent review providers. - A new default `git status` view is added. - A new `commits` view is added to easily navigate all commits and view their descriptions+changes.

@mikeyobrienv Good stuff, Coming from Claude Code, was wondering why Pi tool outputs are so verbose 😥

Tired of verbose output cluttering your terminal? I built pi-tidy-tools to cut the noise and keep things clean. pi install @mobrienv/pi-tidy-tools github.com/mikeyobrien/pi…

@haipingfu Hmm this actually seems pretty useful, might try. Nowadays I’ve just been getting used to massive Claude Code comments to hopefully maintain enough context for future agents.

Introducing Trail! AI coding agents create more than a final diff. They generate prompts, attempts, tool calls, partial patches, checkpoints, test runs, approvals, review notes, and handoffs. But when the work becomes a Git commit, most of that operational context disappears. That’s the problem Trail is built to solve. Trail is a local-first operation database for code and text worktrees. It gives coding agents branch-like memory, transcripts, checkpoints, and rewind—without polluting your active Git branch. Trail does not replace Git. Git remains the publication and synchronization layer: commits, branches, remotes, and PRs. Trail handles the high-frequency local work between commits, preserving what happened, why it happened, and whether it is safe to accept. This makes questions that are difficult to answer with Git alone directly queryable: • Which operation introduced this line? • What did an agent change before the final commit? • What happened during the task? • Is the work blocked by conflicts, pending approvals, a dirty workdir, or missing tests? • Can we rewind to a known-good checkpoint? The core abstraction is a lane: isolated, branch-backed code state plus the context needed to safely complete one task. A lane can record sessions, turns, messages, traces, patches, approvals, test/eval gates, checkpoints, handoffs, and merge readiness. It can stay virtual until a tool actually needs files, or use a materialized or sparse workdir. Under the hood, Trail is written in Rust. A content-addressed graph of operations and worktree roots is the durable source of truth. SQLite powers indexes, lane state, and queues. Prolly maps store paths, files, text, and line order. Stable ChangeId, FileId, and LineId values support provenance and line-aware patching. The same core powers the CLI, a loopback HTTP/OpenAPI interface, an MCP stdio server, and the Rust API. Workspace state remains local under .trail/, with ignore policies, guardrails, approvals, and readiness checks protecting mutations. A basic workflow looks like this: trail init --working-tree trail lane spawn fix-login --from main --materialize=true trail lane record fix-login -m "Fix login validation" trail lane readiness fix-login trail merge-lane fix-login --into main --dry-run Then query the operational history: trail why src/auth.rs:42 trail history src/auth.rs trail transcript fix-login The impact is simple: agent work becomes isolated, explainable, reviewable, recoverable, and safer to merge. Developers can run more agents in parallel without losing control of how the code was produced. Git keeps the accepted history. Trail preserves the operational story behind it. Explore Trail: github.com/crabbuild/trail

@shadcn 👏 Awesome stuff, really nice to work with, incorporated w other Laminar agent updates coming out soon

Today we're releasing a new set of components for building chat interfaces. We've taken the patterns we build every day, rethought the abstractions behind them, and turned them into components you can compose and customize. We're starting with the conversation layer: streaming, scrolling, messages, bubbles, attachments, and markers.

@lumendriada ✅ Thanks for your work!

i honestly need this one. herdr grew much faster than i expected, and because it has no telemetry, most of how people use it is invisible to me. if herdr has been part of your workflow, i’d really appreciate your help.

@ellie_huxtable Have you tried Plannotator or Linear Reviews? Also looking for a better diff viewer…

vibe coded this today so frustrated with GitHub's code review something fable did literally in the background while I worked is much better 100% GPU rendered, native rust. runs at 120fps.

@mitchellh Slowly realizing there are levels to building terminals… 🏆

Ghostty is getting automatic scrollback compression, resulting in 70 to 90% less physical memory usage. It happens incrementally when idle, so it had no measurable effect on IO throughput. I'm not aware of any other mainstream terminal that does this. Demo video below! The gains let us increase the default scrollback limit from 10MB to 50MB, because on average a full scrollback will still compress smaller than the prior limit. More history, for free. ("Unlimited", disk-paged history is on the roadmap too) Let's talk about cool implementation details, cause this was fun. First, the data structure and memory layout ("PageList") I wrote two years ago finally pays off! One of its traits is that screen memory is backed by a linked list of page-aligned, page-sized (or page-multiple-sized) blocks. Because each block is page-aligned and page-sized, we can use madvise to discard its physical backing while keeping the virtual address space reserved. Compressed pages therefore disappear from resident memory, but decompression is still guaranteed because the address space remains valid and we simply fault new pages back in as needed. We use the same trick for our memory pools, too. Unallocated pool pages don't count as resident memory, saving another couple of MB per terminal. This functionality is also available to libghostty-vt consumers via new `ghostty_terminal_compress` APIs. The consumer decides when the appropriate time to compress is and the APIs advise on compressability.

@maxedapps VSCode -> Nvim 🙂

You only need: - herdr + herdr-file-viewer - pi + pi-subagents + pi-web-access - and / or claude code, codex, grok build - decently crafted agent skills which you evolve & improve - including an agent skill (and or custom pi extension) that opens plans, reviews etc as HTML Okay, I also like to add good old VS Code to the mix for when I tweak something by hand (shocking!) or want to dive deeper into the codebase.

Day 2 using @herdrdev! Overall happy with the experience. Tiny nit: It says "idle" even when Claude Code is still working and waiting for something to finish (monitor or background task) which kept throwing me off.

@herdrdev Downloaded an hour ago, positive first impressions 👍 Running inside my own terminal (Ghostty) is a huge plus for me

still leaving your laptop open so the agent doesn't die? still hand-rolling tmux + ssh + notifications? still can't check on it from your phone? you don't have to. try herdr.dev

@plannotator @Nartc1410 Tried this out today, positive first impression 👍 Seems like a product that can get snapped out of existence by Cursor/Claude Code/Linear/etc. (much like many products 😢).

Plannotator v0.21.3 is out Ask AI (plan, review & annotate) - Codex now runs on more reliable app-server Code Review - File-scoped comments + unified comment UX - useful if you create your own guided reviews, ty for the idea @Nartc1410 - Per-file Ask AI chats now open by default - Fixed dead sidebar clicks after switching PRs VS Code extension - Clipboard + keyboard fixes CLI - Subcommand --help fixes

@PrajwalTomar_ Not sure “X was basically a wall”? Unless I’m really missing something important here… Biggest barrier (still) is API cost Easy connection setup already exists with stuff like Composio?

You don't understand how BIG this is. Until now, agents could search the web but X was basically a wall. Real-time posts, trending topics, what people are actually saying right now, none of that was easy to pull into an agent workflow without a complicated API setup. X just shipped a hosted MCP that changes that. Connect Grok, Cursor, or any MCP-compatible tool to the X API with no setup at all. What this actually means if you run agents: → Your content agent can watch what's trending in your niche and surface it before you open your feed → Your research agent can pull real-time reactions to a product launch, not just news articles → Your morning brief now includes live X signal, not just web search I run a content operation across 5 businesses. The bottleneck was always "my agent doesn't know what's happening right now." That just got solved. If you're running agents and not wiring this in this week, you're leaving real-time context on the table.

@raycast Big fan, tiny bug 1 snippet appearing twice in iOS

Does anyone else call their agent "bro"? Building my agent, monitoring with @lmnrai , automatically clusters any issues, biggest one is "User frustration" 😂

If you want to check it out or fork it github.com/kolbeyang/my-a…

My own OpenClaw in 579 lines of TS? OpenClaw's pretty bloated. So I thought, how many lines does it take to get all the features I care about? OpenClaw: 171,348,242 bytes My Agent: 21,560 bytes

Road to self-improving agent? 1. Trace agent with Laminar 2. Give agent Laminar CLI 3. Agent can query anything about previous runs

At Google executives project their importance by how slow they are at approvals. A slowness doom loop. This is why I banned approvals at Vercel. We only have vetos. Want to block something? No problem, speak up. Have nothing to say or taking a week off? Great, it's gonna ship

@NickelanddimeO @JohnPhamous If you’re interested, we built that, open source #transcript-view" target="_blank" rel="nofollow noopener">laminar.sh/docs/platform/…

@JohnPhamous is it possible to drill down on that trace ?

for the eve.dev observability when showing tool calls, we show a snippet of the input

@threepointone @JohnPhamous We gotchu #pii-redaction-for-ai-agent-traces" target="_blank" rel="nofollow noopener">laminar.sh/docs/platform/…

@JohnPhamous do you worry about PII showing up there?