<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Shifu Blog</title><link>https://www.useshifu.com/blog</link><description>Engineering context, carried into your next task.</description><language>en</language><atom:link href="https://www.useshifu.com/blog/feed.xml" rel="self" type="application/rss+xml"/><item><title>Turn an incident lesson into the next task’s test cases</title><link>https://www.useshifu.com/blog/incident-lessons-into-tests</link><guid isPermaLink="true">https://www.useshifu.com/blog/incident-lessons-into-tests</guid><pubDate>Fri, 06 Nov 2026 02:17:22 GMT</pubDate><description>Recover the sequence that produced the failure, state what must remain true, and design a check that would fail if the behavior returned. An incident summary alone is not a regression test.</description></item><item><title>The context handoff between planning, implementation, and review</title><link>https://www.useshifu.com/blog/context-handoff-planning-review</link><guid isPermaLink="true">https://www.useshifu.com/blog/context-handoff-planning-review</guid><pubDate>Fri, 06 Nov 2026 02:17:22 GMT</pubDate><description>Use one compact change record that evolves from problem and constraints to implementation decisions and verification evidence. Give each field an owner and an update trigger so the reviewer receives current understanding rather than a second reconstruction task.</description></item><item><title>What belongs in AGENTS.md—and what should be retrieved per task?</title><link>https://www.useshifu.com/blog/agents-md-or-task-context</link><guid isPermaLink="true">https://www.useshifu.com/blog/agents-md-or-task-context</guid><pubDate>Thu, 05 Nov 2026 04:40:40 GMT</pubDate><description>Put durable, broadly applicable guidance in project instructions. Keep changing decisions, current implementation state, and task-specific evidence in a checked brief with sources. Verify instruction discovery in the tool you actually use.</description></item><item><title>The retrieval optimization we decided not to ship</title><link>https://www.useshifu.com/blog/retrieval-optimization-we-held</link><guid isPermaLink="true">https://www.useshifu.com/blog/retrieval-optimization-we-held</guid><pubDate>Thu, 05 Nov 2026 04:40:40 GMT</pubDate><description>Our candidate found the required records, but so did the baseline. Complete-task estimated cost rose, and one held-out caller answer became less useful. We retained the baseline because the candidate did not pass the gates defined for this study.</description></item><item><title>Give your coding agent a review brief it can use</title><link>https://www.useshifu.com/blog/give-agent-review-brief</link><guid isPermaLink="true">https://www.useshifu.com/blog/give-agent-review-brief</guid><pubDate>Wed, 04 Nov 2026 02:18:03 GMT</pubDate><description>Give the reviewer the problem, current behavior, applicable constraints, sources, and open questions before asking for findings. Then ask it to connect each finding to a concrete failure and a verification step.</description></item><item><title>How we use the context of building Shifu to write about Shifu</title><link>https://www.useshifu.com/blog/building-context-into-blog-posts</link><guid isPermaLink="true">https://www.useshifu.com/blog/building-context-into-blog-posts</guid><pubDate>Wed, 04 Nov 2026 02:18:03 GMT</pubDate><description>We used Shifu to recover relevant product history, checked it against current project evidence, and turned the results into article briefs. The useful part was the connection between a source, an editorial choice, and a claim we could support.</description></item><item><title>Recover architecture decisions before your next AI-assisted change</title><link>https://www.useshifu.com/blog/recover-architecture-decisions</link><guid isPermaLink="true">https://www.useshifu.com/blog/recover-architecture-decisions</guid><pubDate>Tue, 03 Nov 2026 02:12:43 GMT</pubDate><description>Recover a decision as a choice, reason, scope, status, and source. Check what still applies before handing it to an agent; a plausible historical explanation is not the current contract.</description></item><item><title>A good summary can lose the most important detail</title><link>https://www.useshifu.com/blog/summary-important-detail</link><guid isPermaLink="true">https://www.useshifu.com/blog/summary-important-detail</guid><pubDate>Tue, 03 Nov 2026 02:12:43 GMT</pubDate><description>A summary can preserve the topic while losing the condition that changes the next action. Check whether scope, current status, and verification survive—not only whether the output sounds faithful.</description></item><item><title>Measure the cost of rebuilding context before measuring AI productivity</title><link>https://www.useshifu.com/blog/measure-context-reconstruction</link><guid isPermaLink="true">https://www.useshifu.com/blog/measure-context-reconstruction</guid><pubDate>Mon, 02 Nov 2026 02:06:32 GMT</pubDate><description>Measure the whole change, including context preparation and correction after coding. Compare similar work with a clear definition of completion, and keep faster-looking activity separate from a useful outcome.</description></item><item><title>A code review needs the problem statement as well as the diff</title><link>https://www.useshifu.com/blog/review-problem-not-only-diff</link><guid isPermaLink="true">https://www.useshifu.com/blog/review-problem-not-only-diff</guid><pubDate>Mon, 02 Nov 2026 02:06:32 GMT</pubDate><description>A diff shows a change. To judge whether it is the right change, a reviewer also needs the intended behavior, the constraints that must survive, and evidence for the checks that matter.</description></item><item><title>Claude Code memory and context: explain less between tasks</title><link>https://www.useshifu.com/blog/claude-code-memory-context</link><guid isPermaLink="true">https://www.useshifu.com/blog/claude-code-memory-context</guid><pubDate>Sun, 01 Nov 2026 09:09:27 GMT</pubDate><description>Use project instructions for rules that should travel with ordinary tasks, native memory for inspectable learnings, and a task brief for the decisions and exceptions this change depends on. Start by checking what Claude actually loaded.</description></item><item><title>Why we are building a context engine for AI coding</title><link>https://www.useshifu.com/blog/why-context-engine</link><guid isPermaLink="true">https://www.useshifu.com/blog/why-context-engine</guid><pubDate>Sun, 01 Nov 2026 09:09:27 GMT</pubDate><description>We are building Shifu so that decisions, constraints, and reasoning from earlier work can help the next task. The useful unit is an applicable piece of understanding with evidence—not merely a saved conversation.</description></item></channel></rss>
