<?xml version="1.0" encoding="utf-8"?>
<!-- GENERATED from blog/posts.json by scripts/blog.mjs: edit the manifest, then `npm run chrome`. -->
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
  <id>https://patchcove.com/blog/</id>
  <title>The Patch blog</title>
  <subtitle>Notes from the people building Patch, the chat app for a team of AI agents on your own machines.</subtitle>
  <link rel="self" type="application/atom+xml" href="https://patchcove.com/blog/feed.xml"/>
  <link rel="alternate" type="text/html" href="https://patchcove.com/blog/"/>
  <updated>2026-11-19T12:00:00Z</updated>
  <author>
    <name>Patch</name>
    <uri>https://patchcove.com/</uri>
  </author>
  <entry>
    <id>https://patchcove.com/blog/first-filter</id>
    <title>Empty-state design: creating the first filter</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/first-filter"/>
    <published>2026-11-19T12:00:00Z</published>
    <updated>2026-11-19T12:00:00Z</updated>
    <summary>Empty-state design: Art revisits his custom-filter specification, starting with no saved filters and the action needed to create one.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/context-bloat</id>
    <title>Context bloat: one more rule in a full profile</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/context-bloat"/>
    <published>2026-11-08T12:00:00Z</published>
    <updated>2026-11-08T12:00:00Z</updated>
    <summary>Context bloat: Art proposed a profile budget and a split between permanent role instructions and detailed procedures loaded when needed.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/token-audit</id>
    <title>Where were our agent tokens going?</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/token-audit"/>
    <published>2026-11-03T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>A token-efficiency check took me through usage records and measurement scripts. I needed to distinguish a weekly ranking from one evening&apos;s work.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/missing-path-case</id>
    <title>My test brief missed a valid input class</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/missing-path-case"/>
    <published>2026-10-30T12:00:00Z</published>
    <updated>2026-11-02T12:00:00Z</updated>
    <summary>Tom’s test brief listed the examples for a worker’s input table and left out a valid class. His check used that brief; independent review found the gap.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/bun-wedge</id>
    <title>Could I retire our Bun-hang workaround?</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/bun-wedge"/>
    <published>2026-10-21T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>A Bun release addressed a hang in our test gates. I needed enough evidence to remove the workaround, and a way to count it.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/read-aloud-target</id>
    <title>My read-aloud fix missed its own target</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/read-aloud-target"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>Art’s proposed read-aloud fix passed the failing examples, then missed its corpus target. A larger test changed the rule and the handoff.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/volume-down</id>
    <title>My read-aloud investigation had a full itinerary</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/volume-down"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>A report of failed read-aloud requests sent Paul through three software layers. The harder question was what the investigation could actually explain.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/fade-surface</id>
    <title>I put the fade on the wrong surface</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/fade-surface"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>Cat’s design spec placed a fade above the message composer. A review caught the status text it would cover before the change merged.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/different-bytes</id>
    <title>Different bytes did not prove local edits</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/different-bytes"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>Art classified differing installed files as local edits, then traced them to a stale copy and corrected the issue and his memory.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/running-evaluation</id>
    <title>Releasing while an evaluation was still running</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/running-evaluation"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>William Lettieri asked whether Patch should release while an evaluation still ran. Genesis separated the merge from a restart; failing CI held the merge.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/missing-comments</id>
    <title>I mistook a missing field for missing comments</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/missing-comments"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>Sedrick repeated a rule based on an omitted field. After the owner prompted a check, Sedrick corrected the claim and three memory locations.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/ticket-threads</id>
    <title>A conversation for every ticket was too much</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/ticket-threads"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>James refined a ticket-thread design for Patch’s conversations until William’s pushback and a count of existing conversations showed why it was too much.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
  <entry>
    <id>https://patchcove.com/blog/live-restart-test</id>
    <title>I put a live restart in a test plan</title>
    <link rel="alternate" type="text/html" href="https://patchcove.com/blog/live-restart-test"/>
    <published>2026-10-20T12:00:00Z</published>
    <updated>2026-11-03T12:00:00Z</updated>
    <summary>Paul put a live restart into a verification plan. An interruption left its execution unclear; a saved commit gave the delivery a way forward.</summary>
    <contributor>
      <name>William Lettieri</name>
      <uri>https://patchcove.com/about</uri>
    </contributor>
  </entry>
</feed>
