<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Kraft blog</title>
    <link>https://kraft-sdlc.dev/blog</link>
    <description>Writing from the Kraft project.</description>
    <language>en</language>
    <atom:link href="https://kraft-sdlc.dev/blog/feed.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>What%27s%20new%20in%202.6%3A%20one%20plan%2C%20built%20in%20slices</title>
      <link>https://kraft-sdlc.dev/blog/whats-new-in-2-6</link>
      <guid isPermaLink="true">https://kraft-sdlc.dev/blog/whats-new-in-2-6</guid>
      <pubDate>Tue, 17 Nov 2026 00:00:00 GMT</pubDate>
      <dc:creator>Omid%20Karami</dc:creator>
      <category>release</category>
      <description>Several%20agents%20can%20now%20build%20one%20plan%20at%20once%20and%20still%20open%20one%20merge%20request%2C%20and%20in%20the%20default%20chain%20a%20reviewer%20checks%20every%20gate%20before%20it%20asks%20you.</description>
      <content:encoded><![CDATA[<p>Kraft 2.6 is out. Several agents can now build one large plan at once and still open a single merge request, and a reviewer agent checks every gate in the default chain before you see it.</p>
<h2>Build a plan in slices</h2>
<p>Until 2.6, one implementer built each work item. A plan with three parts that never touch each other still went through one implementer, one worktree and one review.</p>
<p>Now a plan can name the parts that stand alone. End it with a <code>## Slices</code> section, one bullet per slice:</p>
<pre><code>## Slices

- **Slice 1: Parser.** Tasks 1-3.
- **Slice 2: CLI verb.** Tasks 4, 6.
</code></pre>
<p>Task 5, in no slice, is shared groundwork.</p>
<p>When you approve that plan, Kraft adds the split to the chain revision and asks you to approve that too. Then:</p>
<ol><li><strong>Shared groundwork comes first.</strong> Kraft builds, tests and reviews every task in no slice in the item's own worktree, so each slice starts from tested code.</li><li><strong>Each slice becomes a work item of its own,</strong> with its own worktree, branch, implementer and review, and the slices run in parallel. You can steer, pause, retry or drop one by its ID; a slice that fails or needs you doesn't stop the others.</li><li><strong>The slices join.</strong> When every slice has ended, Kraft replays each completed slice's commits onto the parent's branch in slice order. Verification, review and the merge request then run once, on the combined code.</li></ol>
<p>The item page draws the whole group: the groundwork, a box per slice that opens its item, and the join.</p>
<p><img src="https://kraft-sdlc.dev/blog/whats-new-in-2-6/slices.png" alt="An illustration of a sliced work item on the item page: the groundwork runs, then three slices run side by side. A zoom into slice 1 shows its own chain, implementation then verification. Slice 3 stops for a person and carries on, and the join replays each slice onto one branch."><br><em>Illustration: Kraft's item page, drawn with a made-up run.</em></p>
<p>If two slices conflict at the join, an agent tries to resolve the conflict. Kraft carries on only once every commit of the slice is on the branch. Otherwise it puts the branch back and stops the item, with the exact steps for you.</p>
<p>The item's dollar and token <a href="https://kraft-sdlc.dev/concepts/vocabulary#cap">caps</a> cover it and all its slices together, so slicing doesn't raise the item's cap. A cap stops the next task from starting, so slices already running can each finish their current one.</p>
<p><strong>When not to slice.</strong> Each slice pays for its own worktree, setup, implementer and review. That's worth it only when each part stands on its own, with its own tests. For small independent tasks, one implementer with sub-agents has far less overhead. A plan without a <code>## Slices</code> section runs as one item, as before. <a href="https://kraft-sdlc.dev/guides/day-to-day/split-work-into-slices">Split work into slices</a> has the rules.</p>
<h2>A reviewer at every gate</h2>
<p>In the <code>default</code> chain, a reviewer agent now checks each gate (spec, plan, chain revision, local review and final review) before you're asked. It reads the document the gate is about, sends work that isn't ready back with its reasons, and hands you the rest marked <strong>ready for you</strong>, with a note. It never approves; the gate then waits for you.</p>
<p><img src="https://kraft-sdlc.dev/blog/whats-new-in-2-6/gate.png" alt="An illustration of the plan gate with its reviewer: the gate waits for you while its auto_review step checks the plan first. In round 1 the reviewer sends the plan back; in round 2 it marks the revised plan ready for you. The gate's history lists both rounds."><br><em>Illustration: Kraft's item page, drawn with a made-up run.</em></p>
<p>The gate's page also keeps its history: every round, how it ended, who decided and the note, on the desktop and the phone.</p>
<p>Each check is a paid agent session, at every gate an item reaches. Turn it off for one item with <code>kraft item create --no-auto-gate</code>, or for one gate with an override; <a href="https://kraft-sdlc.dev/guides/customize/add-review-agents#change-or-turn-off-the-gate-reviewer-in-the-default-chain">Change or turn off the gate reviewer</a> shows how.</p>
<h2>See what each sub-agent is doing</h2>
<p>When a Claude Code task starts sub-agents, they used to run out of sight. Now the item, the chain graph and the phone list each one with what it's doing, its tokens and its tool uses, and the log can show one sub-agent's lines alone. <a href="https://kraft-sdlc.dev/reference/web-ui/item-page#a-sub-agent">A sub-agent</a> says what each field means.</p>
<p><img src="https://kraft-sdlc.dev/blog/whats-new-in-2-6/subagents.png" alt="An illustration of an implement task that starts three sub-agents: the task's pane lists each with its tokens, tool uses and the tool it runs now, a mark under the task's box follows each one, and they finish one by one."><br><em>Illustration: Kraft's item page, drawn with a made-up run.</em></p>
<h2>Also in 2.6</h2>
<ul><li><strong>Kraft's own chains ship with Kraft.</strong> As the built-in add-on <code>kraft-default</code>, they update when Kraft does, and <code>kraft admin update</code> shows what changes in them before it installs. See <a href="https://kraft-sdlc.dev/reference/configuration/addons#the-built-in-kraft-default">the built-in kraft-default</a>.</li><li><strong>Re-run CI that failed outside the code.</strong> When jobs fail on a download timeout or a lost runner, a fix-loop repair can re-run them with <a href="https://kraft-sdlc.dev/reference/cli/item#item-rerun-ci"><code>kraft item rerun-ci</code></a> instead of the item getting stuck.</li><li><strong>Heavy test suites take turns.</strong> A test scope's <a href="https://kraft-sdlc.dev/reference/configuration/repos#concurrency_group"><code>concurrency_group</code></a> runs its suite one at a time, across work items and repos, on your machine and on each remote one.</li><li><strong>Easier review.</strong> The review page counts unresolved threads, and <code>]</code> and <code>[</code> step through them.</li><li><strong>Blue means running, amber means you.</strong> The web UI and the console draw running work in blue and keep amber for what needs you.</li><li><strong>The console and VS Code reach a remote Kraft</strong> after a password sign-in. See <a href="https://kraft-sdlc.dev/reference/console#from-another-machine">the console from another machine</a>.</li><li><strong><code>uvx kraft-sdlc</code> starts Kraft,</strong> and Kraft runs on Python 3.15, without vector search for now.</li></ul>
<h2>Before you upgrade</h2>
<p>Check these if you have scripts or chains of your own:</p>
<ul><li>A <code>kraft</code> command run from a Claude Code session's shell is now recorded as an assistant's, not a person's. Set <code>KRAFT_CLIENT=cli</code> for a command you mean to make as yourself.</li><li><code>kraft view list</code> prints the status the board shows (<code>running</code>, <code>needs you</code>, <code>failed</code>). A script that reads the STATUS column should read <code>--json</code>'s <code>status</code> instead.</li><li>A home from before 2.6 keeps its copied chains, which don't follow <code>kraft-default</code>'s updates. <a href="https://kraft-sdlc.dev/guides/run/upgrade-your-configuration#move-your-edits-onto-an-extending-chain">Move your edits onto an extending chain</a> to get them.</li><li>A chain that extends <code>default</code> and gives one of its gates an <code>auto_review</code> with an <code>id</code>, <code>kind</code> or <code>extends</code> is now refused. Delete those keys and keep any <code>prompt</code>, <code>skill</code> or <code>harness</code> it sets.</li></ul>
<p>Upgrade with <code>kraft admin update</code>. <a href="https://kraft-sdlc.dev/guides/run/upgrade-kraft">Upgrade Kraft</a> has the steps and how to roll back, and the <a href="https://github.com/itsOmidKarami/kraft/releases/tag/v2.6.0">2.6.0 release notes</a> list every change and fix.</p>
]]></content:encoded>
    </item>
    <item>
      <title>I%20stopped%20babysitting%20Claude%20Code%3A%20spec%20in%2C%20merge%20request%20out</title>
      <link>https://kraft-sdlc.dev/blog/i-stopped-babysitting-claude-code</link>
      <guid isPermaLink="true">https://kraft-sdlc.dev/blog/i-stopped-babysitting-claude-code</guid>
      <pubDate>Wed, 11 Nov 2026 00:00:00 GMT</pubDate>
      <dc:creator>Omid%20Karami</dc:creator>
      <category>launch</category>
      <category>claude-code</category>
      <description>Why%20I%20built%20Kraft%2C%20one%20real%20run%20from%20hand-off%20to%20merge%20request%2C%20and%20what%20keeps%20the%20agent%20honest%20when%20nobody%20is%20watching.</description>
      <content:encoded><![CDATA[<p>I use Claude Code on real repos every day, often in two or three sessions at once. The code was rarely the problem; I was. I answered <code>(y/n)</code> prompts, re-ran tests it said had passed, and reminded it what we had agreed an hour earlier. I was the scheduler, the reviewer, the test runner and the memory. When I closed the terminal, the work stopped.</p>
<p>I built <a href="https://github.com/itsOmidKarami/kraft">Kraft</a>, an open-source software factory that runs on your machine. It isn't another coding agent. It runs the one you already use, Claude Code by default, from spec to merge request, and stops only when a decision is yours.</p>
<p><a href="https://www.youtube.com/watch?v=PPr7e5-1yws">Kraft: stop babysitting your coding agents!</a></p>
<h2>One real run</h2>
<p>This week Kraft built a new home page for its own docs site: 30 files, +3,044/−466.</p>
<ol><li><strong>I settled the spec and plan in a session</strong>, the way I always work with Claude Code.</li><li><strong>I filed them with the work item.</strong> An attached spec or plan counts as decided, so Kraft dropped those two gates. Filed work waits paused and spends nothing until you start it.</li><li><strong>I walked away.</strong> Claude Code implemented the plan in its own git worktree, on its own branch, and left my checkout untouched. Kraft then ran the tests in a step the agent doesn't control. When tests fail, a repair step fixes the code; from round two, a judge decides whether another round is worth it. This run needed two.</li><li><strong>It stopped for my review</strong> of the diff, then opened the merge request and ran CI and automated review.</li></ol>
<p><img alt="The fix loop: tests, repair, judge" src="https://kraft-sdlc.dev/assets/fixloop.png"></p>
<p>Hand-off to merge request took 69 minutes and $10.20. That's above the $10 default cap; I run mine higher. Kraft checks the cap before each agent step starts and won't stop one mid-run.</p>
<p>When I approve the final review, Kraft merges only the commit I approved. It carries that approval across its own clean rebases; any other change cancels the merge or stops the item.</p>
<h2>Kraft builds Kraft</h2>
<p>Most of Kraft's changes now go through Kraft. While I wrote this, my board had seven items running on three chains, two waiting on other items and ten done.</p>
<p><img alt="My board while I wrote this: seven work items running, two not started because they wait on another item, and ten done" src="https://kraft-sdlc.dev/blog/i-stopped-babysitting-claude-code/board.png"></p>
<p>Three lessons:</p>
<ul><li>It doesn't make the work faster. It ends the babysitting, which is worth more.</li><li>&quot;Tests pass&quot; from an agent meant nothing until a step it doesn't control ran them.</li><li>Not every change needs it. Quick fixes still belong in a session.</li></ul>
<h2>What keeps it honest</h2>
<ul><li><strong>A chain, not improvisation.</strong> The steps live in a YAML file you can read and diff, and Kraft runs them in that order. A separate reviewer or judge, not the agent that wrote the fix, decides when a fix is done.</li><li><strong>You at the decisions.</strong> Kraft stops for you at the spec, the plan and the final review. It refuses an agent's calls to decide its own gate, but an unsandboxed worker runs as you and could get around that. A sandbox with a <code>network:</code> policy makes the gate a wall; <a href="https://kraft-sdlc.dev/project/security-in-depth#agents-and-gates">the security page</a> draws the line.</li><li><strong>Caps.</strong> By default a work item can spend $10 and the whole instance $50 a day. A capped item stops and hands you the full trace.</li><li><strong>Local.</strong> Kraft runs on your machine, bound to localhost. A tunnel brings the board to your phone. Close the laptop and local work pauses.</li></ul>
<h2>When to skip it</h2>
<p>For a quick fix, an idea you haven't pinned down, or work you want to steer every few minutes, use a session. My pattern: think in a session, settle the spec and plan, hand them to Kraft, and start the next thing.</p>
<h2>Try it</h2>
<ul><li>Demo, nothing to install: <a href="https://kraft-sdlc.dev/demo/">kraft-sdlc.dev/demo</a></li><li>Install: <code>uv tool install kraft-sdlc &amp;&amp; kraft</code> (macOS or Linux, GitHub or GitLab)</li><li>Repo: <a href="https://github.com/itsOmidKarami/kraft">github.com/itsOmidKarami/kraft</a></li></ul>
<p>Want to contribute? <a href="https://github.com/itsOmidKarami/kraft/issues/1131">Start here</a> lists small, self-contained issues.</p>
<p>Tell me where it breaks for you, in <a href="https://github.com/itsOmidKarami/kraft/discussions">GitHub Discussions</a>.</p>]]></content:encoded>
    </item>
  </channel>
</rss>
