clotho-transcript-processor

name	clotho-transcript-processor
description	Use when the user says 'process this transcript', 'extract from this meeting', 'capture transcript', 'here's a transcript', or provides a meeting transcript/notes for extraction. Also triggered when a file is provided with meeting content.

Transcript Processor

Single-transcript processing. The user has a transcript or meeting notes — process it, extract signals, and link everything. This is the manual/ad-hoc version of what the daily debrief does in batch.

The workspace is set automatically. Use clotho_set_workspace if needed.

Step 1: Ingest the transcript

Determine how the user is providing the content:

File path:

clotho_capture(file_path: "<path>", entity_type: "transcript", title: "<meeting name>")

Pasted text:

clotho_create_entity(entity_type: "transcript", title: "<meeting name>", content: "<pasted text>")

Verbal description:

clotho_create_note(title: "<meeting name> - Notes", content: "<what they described>")

If the user doesn't provide a title, infer from the content (first line, meeting subject, attendee names + date).

Step 2: Identify context

Ask (or infer from content):

"Which program does this meeting relate to?"

Present active programs:

clotho_list_entities(entity_type: "Program", status: "active")

If the user specifies a program, read its content to understand the extraction lens:

clotho_read_entity(entity_id: "<program_id>")

If the user says "not sure" or it spans multiple, proceed with general extraction and route later.

Step 3: Create meeting entity (if not already)

If this was a meeting (not just standalone notes), ensure a Meeting entity exists:

clotho_create_entity(entity_type: "meeting", title: "<meeting name>")

Link the transcript to the meeting:

clotho_create_relation(source_id: "<transcript_id>", relation_type: "spawned_from", target_id: "<meeting_id>")

Step 4: Launch extraction

Launch the debrief-processor agent with:

• The single transcript/note entity ID
• The program context (if identified)

The agent will:

• Read program context to understand what signals matter
• Extract speech acts and domain-specific signals
• Create entities with EXTRACTED_FROM relations
• Identify and link people
• Route to programs via BELONGS_TO

Step 5: Review

Present the agent's extraction summary. Ask:

"Does this capture everything? Anything to add or adjust?"

Handle corrections, then:

clotho_sync()

Idempotency

If the user asks to process a transcript that's already been extracted (has EXTRACTED_FROM relations pointing to it), warn:

"This transcript has already been processed. N entities were extracted from it. Want to re-extract (may create duplicates) or skip?"

Check with:

clotho_get_relations(entity_id: "<transcript_id>")

Look for incoming EXTRACTED_FROM relations.