sybill

name	sybill
description	Query your local Sybill transcript store. Ask about specific accounts, deals, action items, or search across calls. Uses ~/sybill-transcripts/ as the source of truth.
license	MIT
allowed-tools	Bash(sqlite3:*, python:*), Read
metadata	{"author":"syftdata","version":"1.0"}

sybill

Answer questions about sales calls using the local transcript store at ~/sybill-transcripts/.

Requires: ~/sybill-transcripts/transcripts.db and ~/sybill-transcripts/calls/{callId}.md files written by sync.py.

Usage

/sybill <natural language question>

Examples

/sybill what's going on with Bloomberg Beta?
/sybill action items from this week
/sybill how did the Tonic demo go?
/sybill calls with Rachel Kirkwood
/sybill deals owned by [e-mail]

Instructions

Step 1: Find relevant calls via SQLite

Use sqlite3 or Python to query ~/sybill-transcripts/transcripts.db.

Common query patterns:

By account name or domain:

SELECT call_id, topic, start_time, account_name, account_domain
FROM calls
WHERE account_name LIKE '%Bloomberg%' OR account_domain LIKE '%bloomberg%'
ORDER BY start_time DESC;

By participant:

SELECT c.call_id, c.topic, c.start_time, c.account_name
FROM calls c
JOIN participants p ON c.call_id = p.call_id
WHERE p.name LIKE '%Rachel%' OR p.email LIKE '%rachel%'
ORDER BY c.start_time DESC;

By date range (this week):

SELECT call_id, topic, start_time, account_name
FROM calls
WHERE start_time >= (strftime('%s', 'now', '-7 days') * 1000)
ORDER BY start_time DESC;

By deal owner:

SELECT call_id, topic, start_time, account_name
FROM calls
WHERE deal_owner_email = '[e-mail]'
ORDER BY start_time DESC;

By meeting type:

SELECT call_id, topic, start_time, account_name
FROM calls
WHERE meeting_type = 'EXTERNAL'
ORDER BY start_time DESC
LIMIT 10;

Step 2: Read relevant markdown files

For each matching call_id, read the full transcript:

~/sybill-transcripts/calls/{call_id}.md

The markdown contains:

• Header with date, duration, account, type, participants
• ## Summary with Outcome, Pain Points, Next Steps sections
• ## Transcript with timestamped speaker lines

Step 3: Answer the question

Based on the user's question, synthesize the relevant information:

"What's going on with X?" - Read all calls for that account, summarize deal status, recent discussion, open questions.

"Action items from this week" - Find calls in the last 7 days, extract Next Steps sections and any tasks mentioned in transcripts.

"How did the [meeting] go?" - Read that specific call, summarize outcome, key moments, sentiment.

"Calls with [person]" - List calls they appeared in, brief summary of each.

Format guidelines

• Lead with a direct answer to the question
• For account summaries: group by call, newest first, with date and key points
• For action items: use a checklist format grouped by account/deal
• For specific call summaries: include outcome, key takeaways, and next steps
• Reference the call date and participants for context
• If no calls are found, say so and suggest running /sync-sybill to pull latest data

Error handling

If transcripts.db does not exist or has no rows:

No transcript data found. Run /sync-sybill (or `python ~/sybill-transcripts/sync.py --all` for a full backfill) to populate the store.