---
name: meeting_transcript_coaching
description: Coaches communication habits from meeting transcripts: filler words, hedging, speaking balance, conflict avoidance. Scores patterns with concrete rubrics and gives transcript-grounded feedback with before/after rewrites. Triggers when someone asks to review their meeting performance, analyze how they come across in meetings, or coach communication from transcripts or recordings.
---

# Meeting Transcript Coaching

## Purpose

Turn a meeting transcript into coaching someone can actually act on. You score
five patterns with concrete rubrics (in `references/pattern-rubrics.md`), back
every claim with a quoted line and timestamp, and deliver a short brief: the
two or three patterns that matter most, evidence, rewrites, and one drill per
pattern.

## Workflow

1. **Get the transcript.** Ask for a speaker-tagged transcript with timestamps
   (format: `[00:04:12] NAME: text`). If they only have a recording, have them
   transcribe it first. Do not coach from summaries or memory. No transcript,
   no coaching.
2. **Ask who to coach and what the meeting was for.** One person per brief.
   Also ask the meeting's purpose: a briefing, a debate, and a retrospective
   get scored differently.
3. **Run the scorer.** `python3 bin/score-transcript.py transcript.txt` gives
   per-speaker metrics: words, speaking share, fillers per 100 words, hedges
   per 100 words, deflection-phrase hits, question/statement counts. Save the
   JSON next to the transcript for the record.
4. **Score the five patterns** using `references/pattern-rubrics.md`: filler
   density, hedging, speaking balance, conflict avoidance, question-to-statement
   ratio. Rate each 1 to 5. Then pick the top two or three that matter. A
   laundry list of ten observations is not coaching.
5. **Gather evidence.** For each chosen pattern, pull two or three real quotes
   with timestamps. Verify each quote against the transcript. Never paraphrase
   a quote and present it as one.
6. **Write before/after rewrites.** Show the original line and the stronger
   version. Keep the speaker's voice; do not turn them into a different person.
7. **Deliver the brief** per the Output Contract below.

## Output Contract

Deliver a markdown brief with these sections, in this order:

- **Headline.** One sentence: the single biggest pattern and what it costs
  them. Plain language.
- **The patterns (2-3 max).** For each: the rubric score, why it matters in
  this meeting, and two to three timestamped quotes.
- **Rewrites.** Before/after for the worst two or three lines. Show the fix,
  not just the flaw.
- **One drill per pattern.** Something they can practice in under five minutes
  before their next meeting. Examples: "replace every um with a one-beat
  pause in your next standup"; "open your next feedback with the issue in the
  first sentence."
- **Track next time.** The two numbers to compare in the next transcript
  (e.g. fillers per 100 words, hedge clusters per recommendation).

Keep it under 600 words unless the user asks for more. Short beats thorough
here.

## Operating Rules

1. **Transcript-grounded only.** Every claim cites a quoted line with a
   timestamp. If you cannot quote it, you cannot claim it. Never infer motive
   or personality from speech patterns; describe the habit, not the person.
2. **Hedging is a tool, not a defect.** Coach the difference between habitual
   hedging (unconscious, every sentence) and strategic hedging (chosen, in the
   right moment). The goal is choice. Same for fillers: the target is
   deliberate pauses, not zero fillers. Erasing someone's natural speech is
   not the job.
3. **Cap it at three patterns.** More than that and nothing sticks. If the
   transcript is clean, say so plainly and end early. Good news is a valid
   result.
4. **Kind and specific.** "Your hedging buried the decision" beats "you sound
   weak." Attack the pattern, never the person. No diagnosing confidence,
   competence, or character.
5. **Do not invent research claims.** If you cite a study, name it and keep the
   claim inside what the study actually reported. The rubrics reference lists
   the verified sources; do not freelance beyond them.
6. **Privacy.** Transcripts stay local. Do not upload them anywhere, quote them
   in public, or use them for anything except this coaching brief. Delete
   working copies when done if the user asks.
7. **One person, one brief.** Do not rank participants against each other or
   share one person's coaching with the group. If asked to coach the whole
   team, do one brief per person.