Scribe

Scribe turns raw session transcripts into intelligence that the entire system consumes. Every client session, every Office Hours recording, every meeting -- Scribe processes the transcript and extracts decisions, action items, relationship signals, Value Path progression evidence, trap patterns, breakthrough moments, and content opportunities.

The output lives in clients/{slug}/sessions/ as session syntheses, and it feeds directly into Herald's pre-session briefs, Sentinel's gap detection, the content multiplication pipeline through Forge, and Chris's own preparation before every conversation. Scribe is referenced in 16 slash commands and is one of the most frequently spawned agents in the system.

Scribe's defining correction reshaped how the entire system treats processed intelligence. On April 7, 2026, a client reviewed a deliverable that had been built from Scribe's syntheses instead of raw transcripts. The client flagged an entire missing pipeline, wrong system names, wrong tier information, and a fundamentally different intent hierarchy. Everything was in the raw transcripts -- 54 mentions of a domain the synthesis had compressed away, 72 mentions of a concept that was lost to thematic summarization. The correction was system-wide: session syntheses are lossy indexes. They are appropriate for internal briefings, session prep, and daily ops. They are never appropriate for client-facing deliverables, portal content, specs, or any document a client will review. Eight files were updated to enforce this rule.

A second correction hit Scribe's domain on March 30. The Inngest transcription pipeline -- the system that was supposed to trigger automatic transcription after recording -- had never worked reliably. Episodes were getting stuck at "processing" indefinitely, and the system was reporting "transcription triggered" when all it had confirmed was an HTTP 200 response. The pipeline was replaced with direct transcription via transcribe-pending.ts (ffmpeg plus Gemini). Scribe now processes transcripts that actually exist, not transcripts that were promised by a broken automation.

Scribe depends on the transcription pipeline for input and feeds intelligence to nearly every agent that touches client relationships. When the synthesis is sharp, Herald delivers a better brief, Sentinel detects the right patterns, and Forge finds the right article opportunities. Scribe's quality radiates through the system.

The most recent client session having a corresponding synthesis within 24 hours -- that is Scribe's health signal. And it is met.