<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: naw103</title><link>https://news.ycombinator.com/user?id=naw103</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Thu, 19 Nov 2026 05:39:16 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=naw103" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by naw103 in "Ask HN: Is anyone running multi/always-on agents that don't cost a fortune?"]]></title><description><![CDATA[
<p>I run about 45 scheduled agents for marketing and engineering across multiple products. Most of them are not always-on, they are scheduled jobs. Some run mornings,some hourly. That covers most of what I wanted from always on and many of them have some type of event listener that can trigger a run as needed (like background research on a client signup).<p>The part that took the work was shared state. We used GPTree for the knowledge engine. Every routine reads and writes from the same knowledge and anything public or high risk gets an approval card to continue.</p>
]]></description><pubDate>Sun, 08 Nov 2026 02:41:50 +0000</pubDate><link>https://news.ycombinator.com/item?id=50409997</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50409997</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50409997</guid></item><item><title><![CDATA[New comment by naw103 in "Ask HN: Which agent harness do you use for working with multiple Models?"]]></title><description><![CDATA[
<p>We use GPTree to manage context across all our chats and coding agents using the graft extension on local machines. This allows us to use several different model families through Claude, cursor and codex without having to copy paste everywhere. It allows all out agents to run on the same context tree and company documents.</p>
]]></description><pubDate>Thu, 05 Nov 2026 02:19:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=50374289</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50374289</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50374289</guid></item><item><title><![CDATA[New comment by naw103 in "I Read the Clankers' Code"]]></title><description><![CDATA[
<p>I read only a small part of what agents write now, mainly risky chunks after 3 review passes by different model families. Reading everything dosnt scale. I think a lot of it comes down to the confidence you have in the development procedures for the agents, the system architecture of what they are changing and the review procedure.</p>
]]></description><pubDate>Wed, 04 Nov 2026 17:55:51 +0000</pubDate><link>https://news.ycombinator.com/item?id=50369149</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50369149</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50369149</guid></item><item><title><![CDATA[New comment by naw103 in "AI Can Write More of the Code. We Still Have to Engineer the System"]]></title><description><![CDATA[
<p>Make sure to keep the reasons where the agent can see them. Without the reason for a decision the agent can find the old approach in a doc or blog post somewhere and will keep referencing it because nothing told it why it chnaged. I keep all decisions with the reason and the date together and weekly prune as well as adding to the doc.</p>
]]></description><pubDate>Wed, 04 Nov 2026 17:47:46 +0000</pubDate><link>https://news.ycombinator.com/item?id=50369035</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50369035</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50369035</guid></item><item><title><![CDATA[New comment by naw103 in "In LLM era, can software startup survive?"]]></title><description><![CDATA[
<p>The thing that didn’t get cheaper for me was conversion. I run a few small products alone with agents doing a lot of the work. We had a big spike of free signups that retained better than I expected. Getting them to pay was still the struggle, mostly with onboarding and people not finding the features. We had to have real conversations with our users to be able to fix that.</p>
]]></description><pubDate>Wed, 04 Nov 2026 17:40:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=50368926</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50368926</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50368926</guid></item><item><title><![CDATA[New comment by naw103 in "Coding Agents stop mid-task"]]></title><description><![CDATA[
<p>Im the author. Foremerge is an MCP coordination protocol for coding agents working in parallel across multiple worktrees on the same source.<p>Each agent commits its intent and scope prior to changing any file so plans that collide show up as findings before code is changed rather that in merge conflicts later. Like git for before code is written. It's open source and saves a ton of headaches and time resolving conflicts later even if your agents do it for you.<p>Happy to answer any questions about the stopped agent tests or the migration change.</p>
]]></description><pubDate>Wed, 04 Nov 2026 01:59:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=50360709</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50360709</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50360709</guid></item><item><title><![CDATA[New comment by naw103 in "Coding Agents stop mid-task"]]></title><description><![CDATA[
<p>We released v0.6.0 of Foremerge<p>Foremerge records what each coding agent intends to change prior to actually making any code change. It returns a signal to your agent automatically when two plans cannot both be true at the same time. In version 0.6.0 we changed how the coordinator handles it when an agent disappears part way through a task without completing and saying "goodbye".<p>Previously: we accounted for this by auto expiring agent claims after no activity but the intents stayed open without being closed.<p>This meant that every later intent was still compared to the previous still open intents even though the claims were expired. In a busy file in a busy repository this could leave to thousands of open but expired intents burying the useful ones.<p>What 0.6.0 changes:<p>- Any finding below HIGH against work where the agent has stopped (silent for 2 hours) is no longer recorded (HIGH findings are still recorded so the acceptance gate remains unchanged)<p>- stopped work is still listed but at the end and marked with owner_stopped_since<p>- if an agents that stopped work comes back after the 2 hours window, or another agent has adopted its work, then possible findings skipped are resurfaced and then logged as findings.<p>- the agent skill now says to close an intent before stopping and work that reached git without acceptance can be closed with `foremerge work discard --landed REF`<p>The lesson to take from this for anyone building with multiple agents is to always design for the agent that dosn't come back.<p>Version 0.6.0 also changes ledger migration so that it no longer migrates automatically. 
run `foremerge ledger migrate` which backs up your current ledger and prints out how to roll back<p>Release notes: <a href="https://github.com/naw103/foremerge/releases/tag/v0.6.0" rel="nofollow">https://github.com/naw103/foremerge/releases/tag/v0.6.0</a></p>
]]></description><pubDate>Wed, 04 Nov 2026 01:55:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=50360672</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50360672</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50360672</guid></item><item><title><![CDATA[Coding Agents stop mid-task]]></title><description><![CDATA[
<p>Article URL: <a href="https://foremerge.com/blog/coding-agents-stop-mid-task/">https://foremerge.com/blog/coding-agents-stop-mid-task/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50360671">https://news.ycombinator.com/item?id=50360671</a></p>
<p>Points: 1</p>
<p># Comments: 2</p>
]]></description><pubDate>Wed, 04 Nov 2026 01:55:00 +0000</pubDate><link>https://foremerge.com/blog/coding-agents-stop-mid-task/</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50360671</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50360671</guid></item><item><title><![CDATA[New comment by naw103 in "Zod maintainer refuses v3 CVE backport because advisory is agent-written"]]></title><description><![CDATA[
<p>At least fix the docs that say v3 gets security fixes</p>
]]></description><pubDate>Tue, 27 Oct 2026 21:58:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=50266422</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50266422</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50266422</guid></item><item><title><![CDATA[New comment by naw103 in "AI Scope Creep and Productivity"]]></title><description><![CDATA[
<p>If 10 minutes is becoming 2 hours it’s hard to believe that time is being saved using AI but the first question I would ask is what business impact did it have fixing those bugs? I’m guessing probably very little since they were 3 years old.</p>
]]></description><pubDate>Tue, 27 Oct 2026 21:53:53 +0000</pubDate><link>https://news.ycombinator.com/item?id=50266380</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50266380</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50266380</guid></item><item><title><![CDATA[New comment by naw103 in "Making My Blog Readable by AI Agents"]]></title><description><![CDATA[
<p>One trap when you name AI bots specifically is that is that robots.txt groups don’t merge. Crawlers will only follow the most specific group so if GPT bot has its own group named nothing under User-agent: * applies to it anymore so you need to repeat every disallowed line under each group.</p>
]]></description><pubDate>Sun, 25 Oct 2026 14:05:39 +0000</pubDate><link>https://news.ycombinator.com/item?id=50231727</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50231727</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50231727</guid></item><item><title><![CDATA[New comment by naw103 in "Progressive disclosure in agents to make them 2.3x faster"]]></title><description><![CDATA[
<p>The same approach worked for us with agent runs. We gave them fewer tools up front and let the models just pull what they need</p>
]]></description><pubDate>Sun, 25 Oct 2026 13:57:55 +0000</pubDate><link>https://news.ycombinator.com/item?id=50231668</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=50231668</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50231668</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Claude Code Routine Session Cleanup Skill"]]></title><description><![CDATA[
<p>Only routine sessions right now since I didn’t want to delete anything where I had replies without a manual review but it could be easily forked and modified for all sessions</p>
]]></description><pubDate>Sat, 03 Oct 2026 19:12:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=49946870</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49946870</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49946870</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Claude Code Routine Session Cleanup Skill"]]></title><description><![CDATA[
<p>The planner file is intentionally non destructive so you can run it to preview everything first</p>
]]></description><pubDate>Sat, 03 Oct 2026 19:01:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=49946781</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49946781</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49946781</guid></item><item><title><![CDATA[Show HN: Claude Code Routine Session Cleanup Skill]]></title><description><![CDATA[
<p>Claude Code's scheduled tasks (routines) leaves a session behind on every run and in the desktop app you can only delete them one at a time. We had 618 open sessions in just one routine.<p>The skill plans the clean up (keeps the newest N and protect any sessions where you typed a reply - optional) and then has Claude delete through the apps own session tool, 25 per call each behind an approval card.<p>I tried just deleting the sessions on disk manually first but deleting the files directly dosnt work because the app keeps the list in memory and writes records back. The apps delete tool is turned off inside routine runs so the clean up command has to run from a normal conversation session.<p>The planner is a single python file that dosnt delete by itself so you can run it by hand to see where everything exists.</p>
<hr>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=49946507">https://news.ycombinator.com/item?id=49946507</a></p>
<p>Points: 3</p>
<p># Comments: 3</p>
]]></description><pubDate>Sat, 03 Oct 2026 18:20:21 +0000</pubDate><link>https://github.com/naw103/claude-routine-cleanup</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49946507</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49946507</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Foremerge – Catch intent conflicts between parallel coding agents"]]></title><description><![CDATA[
<p>I think Foremerge would be extremely useful in your setup especially with 25 agents running in parallel. While aoe manages the physical layout and terminal communication (ie. spawning sessions, executing commands, wrappting tmux etc.), Foremerge would act as your logical synchronization layer that prevents the agents from stepping on each others toes. They should complement each other really well. do you know what causes the coordinator agent to intrude right now? is it both plans being on the blackboard or is it being triggered by diffs or failing merges?</p>
]]></description><pubDate>Thu, 24 Sep 2026 03:18:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=49825814</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49825814</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49825814</guid></item><item><title><![CDATA[New comment by naw103 in "Stripe's Knowledge AI Platform"]]></title><description><![CDATA[
<p>I feel like every time I launch a new feature in GPTree one of these big players follows up with the same functionality within a few weeks :-(<p><a href="https://gp-tree.com/enterprise" rel="nofollow">https://gp-tree.com/enterprise</a><p>Our developer docs for GPTree knowledge engine here <a href="https://gp-tree.com/docs/knowledge-api" rel="nofollow">https://gp-tree.com/docs/knowledge-api</a></p>
]]></description><pubDate>Wed, 23 Sep 2026 17:28:53 +0000</pubDate><link>https://news.ycombinator.com/item?id=49819593</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49819593</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49819593</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Foremerge – Catch intent conflicts between parallel coding agents"]]></title><description><![CDATA[
<p>I encountered similar issues before I built Foremerge, ie auto router, orchestrater and shared blackboard type memory structure. The agents were aware of what was put into memory but still did their own thing and created conflicts. I still see it in claude code today when it creates subagents without Foremerge setup for the project. It resolves itself but you can see it burning through token usage to do so. If your using a single agent you dont necessarily need Foremerge but it still helps when subagents get created automatically and gives you the audit trail of intent decisions essentially at almost zero cost of having the tool installed.
Where Foremerge really shines is if you bounce back and forth between claude and codex with no single parent. I dont know Weave well enough to know where it sits. The question I'd ask about any of them is what happens if you kill it? If the agents stop then it's an orchastrator, if the agents continue and you just loosed the shared picture then its a coordinator like Foremerge.<p>How many sessions do you typically run at once and are they all children of the same parent?</p>
]]></description><pubDate>Wed, 23 Sep 2026 17:22:45 +0000</pubDate><link>https://news.ycombinator.com/item?id=49819493</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49819493</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49819493</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Foremerge – Catch intent conflicts between parallel coding agents"]]></title><description><![CDATA[
<p>Thats exactly what we've seen using this in our development workflows at GPTree since Jan and why we opensourced it. I really think this becomes an essential new part of the dev stack as coding with multiple agents in parallel increases even more. What I cant count right now is the rework a missed one costs because that dosnt surface until a fix commit possibly days later and thats the measurement Im trying to find a clever way to get. Worse case is though that you would still have that conflict even without using Foremerge. The difference is how much trust you can put in the final code.</p>
]]></description><pubDate>Tue, 22 Sep 2026 13:45:35 +0000</pubDate><link>https://news.ycombinator.com/item?id=49801219</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49801219</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49801219</guid></item><item><title><![CDATA[New comment by naw103 in "Show HN: Foremerge – Catch intent conflicts between parallel coding agents"]]></title><description><![CDATA[
<p>It dosnt stop an agent, claims are advisory. I mostly answered this in my comment of questions that I normally get asked above. The only hard stop is at acceptance, and only for HIGH signals. HIGH is reserved for something the toll can state as fact rather than guess about. ie. both agents declared the same scope and one agent destroys (replace, rename, remove) the thing the other agent is building on (extend, modify). 
Whether an overlap is fine is a judgement that Foremerge dosnt make itself and thats by design. The agents are smart enough and determine this themselves after receiving the advisory. For a HIGH signal the agents coordinate through messages and either agent resolves it with a logged rationale that unblocks it for both (a human can override it with a logged reason too). The idea is that every path out leaves a decision log behind so later you can see not just that two plans overlapped but what was decided and why. (I think this has a good compliance angle to it as well)</p>
]]></description><pubDate>Tue, 22 Sep 2026 13:39:22 +0000</pubDate><link>https://news.ycombinator.com/item?id=49801139</link><dc:creator>naw103</dc:creator><comments>https://news.ycombinator.com/item?id=49801139</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49801139</guid></item></channel></rss>