Self-coaching from meetings: where am I actually avoiding hard things

Ask Claude where you've been ducking confrontation, dominating airtime, or rushing past concerns. Specific moments. Named patterns. Better alternatives. All grounded in what you actually said this week.

What you'll have when you're done

A set of four coaching prompts you run against your own meeting transcripts once a week. Each prompt asks Claude to look across the last seven days and surface a specific leadership pattern with verbatim quotes and the meeting it came from. You read the output Sunday night with a coffee. You flinch at least once. You walk into Monday knowing the exact pattern to fix and the exact moment you'll get to fix it.

No therapist. No leadership coach. No 360 review. The interface is claude code in a terminal. Your meeting library is the folder of transcripts [Granola][1] writes to your laptop, set up via [Granola → markdown][2]. The coach is your own week's conversations, read back to you with the parts you'd rather skip highlighted.

The leader I think I am vs the leader the transcripts show

I'd told myself for years that I was a good listener. Then I started running coaching prompts against my own 1:1s. Then I read what came back. Then I read it again because I was sure Claude had it wrong. Then I opened the source meetings and watched myself interrupt my VP three times in eight minutes, dismiss a concern Diego raised by reframing it as "an opportunity," and end a sync without naming the one thing both of us knew was the actual problem.

I had been doing this for years. The whole team had noticed. Nobody had ever told me.

That's the thing about leadership patterns. The cost is real and the feedback loop is broken. The people on your team learned a decade ago that telling the founder he interrupts doesn't go well. So they stop telling you. So you keep doing it. So the people who could push back gradually stop bringing the hard stuff. So you start running a company that looks like it's working from the inside and is leaking decisions, candor, and conviction from the outside.

The transcripts are the only honest mirror. Granola was already capturing everything. The meeting library was already on my laptop. The coaching workflow is what turned that vault from a memory tool into a self-coaching tool. The output is uncomfortable on purpose. The discomfort is the work.

What you need first

• The Granola → markdown pipeline already running. This workflow runs on the meeting library the pipeline produces. If you haven't set it up yet, do [Granola → markdown][3] first. Thirty minutes. Comes back when done.
• Claude Code installed and pointed at the folder your Granola transcripts land in. Free tier is sufficient.
• At least two weeks of meetings in your meeting library. A single week is not enough surface area to surface a pattern. Two weeks of 1:1s, leadership syncs, and customer calls is the floor. Four weeks is better.
• One uncomfortable hour blocked on a Sunday. This workflow only works if you actually look at the output. Block it now, before you write any prompts.

Step-by-step

Step 1Set the posture before you write any prompts

This is the step that decides whether the workflow lands. The output of these prompts is designed to be blunt. If you read it defensively, you'll dismiss it, refine the prompt to be gentler, and lose the workflow. So before you run anything, agree with yourself on three things.

1. Claude isn't right about everything. Your meeting library is finite. The patterns it surfaces are real moments; the interpretation might be off. Read the source quote, not just the verdict. If the quote doesn't support the pattern, push back.
2. One uncomfortable answer is a win, not a failure. The point isn't a clean scorecard. The point is one specific moment, named, that you can fix in next week's meetings. If a run surfaces nothing uncomfortable, the prompt is too gentle. Tighten it.
3. Nobody sees this but you. The output file lives on your laptop. No coach reads it. No board sees it. The privacy is the unlock. You can be more honest about what you're avoiding when nobody is grading you.

Skip this step and the workflow becomes a thing you set up once and quietly stop running by week three.

Step 2Run the confrontation-avoidance prompt

The first prompt is the load-bearing one. It looks across every meeting in the last thirty days and finds the moments you ducked a hard topic.

Open Claude Code in your notes folder:

cd ~/notes
claude

Paste this prompt:

Look across every team meeting and 1:1 in ~/notes/granola/ from the last 30 days.

Find three moments where I dodged a hard topic, deferred a decision, or ended
a conversation without naming a problem I knew was there. Be specific.

For each moment return:

- **Meeting:** filename and date
- **What I did:** one sentence describing the duck
- **Verbatim quote:** the exact words I used to defer, redirect, or close out
- **What was actually in the room:** one sentence on the topic I avoided
- **What I could have said instead:** one direct alternative, in my voice

Rules:
- Quote me verbatim. If I said "let's circle back," show me "let's circle back."
- Don't soften the assessment. If I dodged, say I dodged.
- Skip the moments where deferring was the right call (we genuinely needed
  more information). Only surface the ones where the information was already
  in the room and I deferred anyway.

Read the output. Open one of the cited meetings. Find the moment Claude flagged. The first uncomfortable answer is the workflow. That's the signal you can trust the prompt and your meeting library.

Step 3Run the airtime-dominance prompt

Next prompt looks at your 1:1s and estimates how much of the conversation was yours versus theirs.

Across every 1:1 I had in ~/notes/granola/ in the last 30 days, estimate the
talk-time ratio (mine vs theirs). For each 1:1 return:

- **Person:** name
- **Meeting:** filename
- **Estimated ratio:** mine vs theirs, expressed as a percentage split
- **Flag:** any 1:1 where I spoke more than 60% of the time
- **One moment:** quote one specific exchange where I could have made space
  for them and didn't (interrupted, talked over, finished their sentence,
  or filled silence I should have left open)
- **What it cost:** one sentence on what they probably didn't say because
  of the moment above

Be blunt. Don't soften the flags. The whole point is to surface the pattern.

The first time I ran this, three of five 1:1s came back flagged. The "what it cost" line on one of them named a concern my Head of Sales had been sitting on for a month. I'd been filling the silence she was using to work up to saying it.

Step 4Run the rushed-concerns prompt

Third prompt finds moments where someone raised a concern and you steamrolled, dismissed, or reframed it.

Read every meeting in ~/notes/granola/ from the last 30 days.

Find moments where someone (a teammate, a customer, a candidate) raised a
concern and I dismissed, rushed, or reframed it instead of staying with it.

For each moment return:

- **Meeting:** filename
- **Who raised the concern:** name
- **Their verbatim concern:** the exact sentence(s) they used
- **What I did:** one sentence describing how I handled it (dismissed,
  reframed as an opportunity, jumped to a solution, moved on)
- **What I said:** my verbatim response
- **What staying with it would have sounded like:** one alternative response
  that acknowledges the concern before solving it

Skip the moments where the concern was genuinely handled in the next two
minutes of conversation. Only surface the ones I closed without resolving.

The "reframed as an opportunity" line is the one that hurts. Most CEOs do it constantly. The reframe is a coping mechanism for handling concerns at speed; the cost is the concern goes underground and shows up in week three as a resignation, a churned customer, or a board pre-read full of surprises.

Step 5Run the decision-deferral prompt

Fourth prompt is the one that surfaces decisions you didn't make.

Read every meeting in ~/notes/granola/ from the last 30 days.

Find every decision I deferred, kicked, or said I'd "think about" and then
didn't return to in a later meeting.

For each return:

- **Meeting:** filename and date
- **The decision in front of me:** one sentence
- **My verbatim deferral:** the exact phrase I used ("let me think about it,"
  "let's revisit next week," "I want to sleep on it," etc.)
- **Whether I returned to it:** check meetings since for follow-up. Yes / no /
  partial. If yes, when. If no, flag it.
- **Cost of the delay:** one sentence on what the team or customer was waiting
  on while I was deferring

Be specific. "Decided not to" is a decision and doesn't count as deferral.
"Said I'd think about it and never brought it back up" does count.

This one is the workflow that's saved me the most. Two of my biggest mistakes as a CEO were decisions I deferred for three months because I didn't have to make them in any specific meeting. Nobody was holding me to a deadline because nobody knew I was deferring. The prompt closes that loop.

Step 6Save the four prompts and schedule the weekly run

The four prompts above are the gallery. Save them once so you can re-run them with one command.

mkdir -p ~/.claude/prompts
touch ~/.claude/prompts/self-coaching.md

Paste all four prompts into that file with clear headers. Each Sunday night, open Claude Code in your notes folder and run:

Run the four coaching prompts in ~/.claude/prompts/self-coaching.md against
my meetings from the last 7 days. Write the output to
~/notes/self-coaching-{date}.md.

You get one file per week. Read it Sunday night. Mark the one pattern you're going to fix in the upcoming week. That's the only deliverable.

If you want to automate it further, get Claude to write a [Routine][4] that runs the four prompts every Sunday at 6pm and emails you the report. Twenty minutes to set up. After that, the file lands in your inbox and you just read it.

How you'll know it's working

The signal is the first time you flinch.

You'll read a line like "you spoke 71% of your 1:1 with Priya and the moment you interrupted her was when she was about to raise the staffing problem on the platform team." You'll feel called out. You'll want to argue with Claude. You'll open the source meeting and find the moment. The interruption will be there. The concern she didn't get to raise will be there.

That moment is the workflow.

A secondary signal: the Monday after your first uncomfortable run, you'll notice yourself catching the pattern in real time. Mid-1:1 you'll realize you're about to interrupt and you won't. The 1:1 will go ten seconds longer than usual. They'll say something you needed to hear. The fix lives in the noticing, and the noticing comes from the discomfort of last night's read.

When it breaks

• Claude is too gentle. The default tone is supportive. If the first run comes back with "you had a thoughtful conversation with Diego where both perspectives were heard," the prompt is too soft. Add to the prompt: "Be blunt. I'm not looking for affirmation. I'm looking for the one moment in each meeting I'd be embarrassed to watch on video." Re-run. The output sharpens immediately.

• You start refining the prompts to feel better. This is the silent failure mode. After a few uncomfortable runs, the temptation is to soften the rules ("only flag interruptions that lasted more than 10 seconds"). Resist. The friction is the workflow. If the output feels less uncomfortable each week, either you actually changed (great) or the prompt got gentler (not great). Spot-check by re-running the original prompt on the same week.

• Selection bias toward your bad moments. The prompts ask for failures. The output will look one-sided because it was asked to be one-sided. The fix is to occasionally run a balancing prompt: "Find three moments in the last 30 days where I handled something well. Specific quotes." Read both files together. The balance keeps the workflow sustainable.

• You read the output and don't act on it. The workflow only works if the noticing converts into a change in next week's meetings. If three weeks go by and the same pattern keeps surfacing, the bottleneck isn't the prompt. The bottleneck is that you read the output as information instead of as instruction. Pick one pattern. Fix it for one week. Then move to the next.

• Your meeting library doesn't have the right kind of meeting. Coaching prompts surface patterns in interactions; if your meeting library is mostly solo work sessions or recorded talks, there's nothing to find. The workflow needs 1:1s, leadership syncs, customer calls, and team meetings. If those aren't being recorded, fix that first.

Where this fits in your harness

This is the leadership-coaching layer of your [harness][5]. The [Granola → markdown pipeline][6] made every conversation searchable. [The query layer][7] lets you ask any question of your meeting library. Self-coaching is a specific, repeated, uncomfortable version of that query: four prompts, every week, aimed at the patterns you can't see from inside the meeting.

The siblings most relevant to this workflow:

• [The commitment ledger][8] · the accountability layer for what you owe and what's owed to you. Self-coaching closes a different leak: the things you didn't say, the questions you didn't ask, the decisions you didn't make. Together they cover both sides of the pattern.
• [Ask your meeting history anything][7] · the query muscle this workflow depends on. Build that one first if you haven't. The coaching prompts are paste-ready versions of the same shape.

See the [Granola pillar][9] for the full pipeline and the other workflows that compound on top.

[1]: /workflows/what-is-granola [2]: /workflows/granola-to-markdown [3]: /workflows/granola-to-markdown [4]: /workflows/claude-routines [5]: /workflows/what-is-a-harness [6]: /workflows/granola-to-markdown [7]: /workflows/ask-your-meeting-history [8]: /workflows/commitment-ledger [9]: /blog/granola-for-ceos-highest-roi-ai-install