meeting-transcription

Meeting Transcription Pipeline

A pipeline that turns meeting recordings into clean, structured Markdown transcripts with named speakers. It is context-aware: each context (e.g. main, customer-x) has its own voice library, language default, and output routing, so the same system serves internal standups and customer calls in different languages without reconfiguration.

This skill is both the knowledge base (how the system works) and the operations manual (how to change it). The canonical scripts live in scripts/ here and are deployed to the worker host's ~/transcribe-pipeline/. It is the reference implementation of the bring-your-own-worker contract in docs/transcription-worker.md — /debrief talks to it only through scripts/debrief_sync.sh (pull / push).

Configuration — the yaml wiring

Three config surfaces, same pattern as every other capability:

1. bridge-config.yaml — the integration block (is the capability on, which script, which contexts — registration + routing):

integrations:
  transcription:
    enabled: true
    sync_script: "skills/meeting-transcription/scripts/debrief_sync.sh"
    default_context: main        # context for audio handed off without an explicit one
    contexts:
      main:                      # → workflow/contexts/main.yaml
        imports: work/imports    # where this context's transcripts land on pull

BRIDGE_IMPORTS / TRANSCRIBE_CONTEXTS env-override the imports dir + context list. /debrief degrades gracefully to its manual path when the block is absent (see the contract doc).

2. infra/transcriptions/topology.yaml — placement (WHERE the pipeline runs):

mode: remote                   # local | remote — THE topology switch (env TRANSCRIBE_MODE overrides)
worker:                        # consulted only when mode == remote
  host: worker-host            # SSH/Tailscale alias — see infra/remotes/
  launchd_label: com.openbridge.transcribe-worker
# local:                       # consulted only when mode == local; all keys optional
#   inbox_dir:       ~/transcribe-inbox        # defaults = the worker's own conventions
#   transcripts_dir: ~/Transcripts
#   library_dir:     ~/transcribe-pipeline/speaker-library
#   contexts_dir:    ~/transcribe-pipeline/contexts