<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Agent Workflows knowledge base</title><description>Learned experience from building and running agent workflows.</description><link>https://workflows.doruk.uk/</link><item><title>Diagnose a run in the Langfuse UI</title><link>https://workflows.doruk.uk/knowledge-base/observability/diagnose-in-langfuse-ui/</link><guid isPermaLink="true">https://workflows.doruk.uk/knowledge-base/observability/diagnose-in-langfuse-ui/</guid><description>A step-by-step manual, recorded on a real slow-run question on 2026-10-05. The run is CI run ███████████ of the ██████████-ai/██████████-Frontend-V2 trial workflow: ████████ on v2-preview, the self-hosted Mac, action revision fa890e9. The screenshots come from Langfuse 4.50.0, with the user identifier redacted. The same steps work for any traced run. Observability explains what the spans mean. Diagnose a run with the CLI gives exact totals for the same run.</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>langfuse</category><category>observability</category><category>claude-code</category></item><item><title>Diagnose a run with the CLI</title><link>https://workflows.doruk.uk/knowledge-base/observability/diagnose-with-cli/</link><guid isPermaLink="true">https://workflows.doruk.uk/knowledge-base/observability/diagnose-with-cli/</guid><description>This guide works through the same CI run as the UI manual: ████████, CI run ███████████ of the ██████████-ai/██████████-Frontend-V2 trial workflow, recorded on 2026-10-05. Use it when you need exact totals, a comparison between runs, or an agent to do the diagnosis. Every command and output below is from that session. Long paths are shortened to ….</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>langfuse</category><category>observability</category><category>claude-code</category></item><item><title>Effort level comparison: ████████ (2026-10-05)</title><link>https://workflows.doruk.uk/knowledge-base/observability/evidence/2026-10-05-effort-level-att-5933/</link><guid isPermaLink="true">https://workflows.doruk.uk/knowledge-base/observability/evidence/2026-10-05-effort-level-att-5933/</guid><description>This is the evidence behind #116: which reasoning effort the verification agent should run at. The raw per-run numbers are in 2026-10-05-effort-level-att-5933.json. They were read from Langfuse with the queries in Diagnose a run with the CLI.</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>langfuse</category><category>observability</category><category>claude-code</category></item><item><title>Evidence</title><link>https://workflows.doruk.uk/knowledge-base/observability/evidence/README/</link><guid isPermaLink="true">https://workflows.doruk.uk/knowledge-base/observability/evidence/README/</guid><description>This folder holds measurements that back decisions about walkthrough speed and behavior. Each entry is a dated write-up with its raw data next to it.</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>langfuse</category><category>observability</category><category>claude-code</category></item><item><title>Observability</title><link>https://workflows.doruk.uk/knowledge-base/observability/README/</link><guid isPermaLink="true">https://workflows.doruk.uk/knowledge-base/observability/README/</guid><description>This folder explains how to see where a walkthrough&apos;s time goes. The verification agent can export a timed trace of every model request and tool call to a self-hosted Langfuse. People and agents then read that trace in the Langfuse UI or with the Langfuse CLI. ADR-0020 records the decision and the bake-off behind it.</description><pubDate>Mon, 05 Oct 2026 00:00:00 GMT</pubDate><category>langfuse</category><category>observability</category><category>claude-code</category></item></channel></rss>