Conflict Avoidance Analysis
Meetings analyzed: 2 (Sept 15 weekly sync, Sept 29 1:1 with Priya). Both are short, about 2 minutes each, so these are patterns from a small sample. I found 5 instances of avoidance, all in the weekly sync. The 1:1 showed none.
Instances of avoidance (Weekly Sync, 9/15)
1. Hedging on the schedule slip (00:00:41)
"Right, right. So, I mean, maybe we could, like, potentially look at the timeline again? If you think that makes sense. But whatever works for you."
Priya had just said the project was two weeks behind. You stacked hedges ("maybe", "like", "potentially", "if you think"). Then you handed the decision back with "whatever works for you." That gave her no direction on a real problem.
Better: "Two weeks behind is a problem. Let's decide now: cut scope or add someone?"
2. Deferring her solution (00:01:10)
Priya: "I think we need to cut scope or add a person."
Alex: "Yeah, that's a good thought. Let's, um, circle back on that."
She gave you two concrete options. You praised the idea but didn't decide anything or commit to a time. That is agreeing without commitment.
Better: "Both are on the table. I'll come back with a decision by Wednesday."
3. Glossing over the vendor risk (00:01:35)
Marcus: "...the vendor hasn't confirmed the cutover date."
Alex: "Okay, sounds fine. I'm sure it'll work out. Let's move on to hiring."
Marcus raised a risk and you said "sounds fine" and changed the subject. Marcus had to bring it up again at 02:20, when he said it could slip the whole quarter.
4. Shutting down Priya's request (00:01:58)
Priya: "Can we spend two minutes on the onboarding timeline before we move on?"
Alex: "Sure, sure, yeah, we can take that offline. Sorry, I'm just conscious of time."
This is the clearest example. Priya asked directly for time on the issue. You said "sure" but then declined, with an apology and a time-pressure reason. The meeting had time for the vendor discussion right afterward. The timeline problem never got resolved in the room.
5. Dropping the topic (whole meeting)
The onboarding delay was never resolved. It was raised at 00:25, deflected at 00:41 and 01:10, and deferred at 01:58. The agenda moved on each time it came up.
What changed afterward
- Onboarding: Priya was still stressed two weeks later (9/29). She said she had redone the flow three times and was "nowhere near" the deadline. The unresolved sync probably contributed to that.
- Vendor risk: Marcus escalated, and you handled that well (02:31–02:47). You asked what he thought, accepted his suggestion, and assigned the email with a deadline. It took a second push to get there.
Where you did it well
1:1 with Priya (9/29):
- You asked an open question: "What's getting in the way, specifically?" (00:34)
- You checked your understanding: "Did I get that right?" (01:02)
- You committed to a direct action: "I'll talk to Dana today... I'll tell her directly that further changes will move the date." (01:18) That is a conflict you were willing to take on.
- You gave specific praise (01:40).
Compare that with the sync. In the 1:1, your language was direct and committed. In the group meeting, it was hedged and you deferred.
Pattern and recommendations
Pattern: You avoid conflict in group settings when the news is bad and someone is waiting for you to decide. You do fine one-on-one. The trigger seems to be a public problem that needs a decision. Time pressure ("I'm conscious of time") is your usual exit.
- Treat "we're behind" as a decision point. Don't say "let's circle back" unless you name when.
- Drop the hedge stack. Cut "maybe", "like", "potentially" and "whatever works for you." Make a recommendation, even a tentative one.
- Honor direct requests for time. If someone asks for two minutes, give them two minutes, or say "no, because X, and we'll talk Thursday at 2."
- Don't say "sounds fine" about a risk without asking one follow-up question.
- Bring the 1:1 directness to the group. The sentence you used with Priya, "I'll tell her directly that further changes will move the date," is the tone you want.
Two meetings isn't much data. If you add more transcripts, I can check whether this holds across other group meetings, and also count filler words and speaking ratios.
Real run recorded with claude-code / claude-sonnet-5-5. Run on a sample folder with 2 invented meeting transcripts, which we wrote ourselves; all names are fictional. Tool-call lines show only the first line of each command. The model output is unmodified.
できること
会議の文字起こしを読み、あなたのコミュニケーションについて、引用とタイムスタンプを根拠にした個人向けのフィードバックを返します。
仕組み
- フォルダー内の文字起こしファイル(txt、md、vtt、srt、docx)を探し、話者ラベルとタイムスタンプを確認します。
- 知りたいことが示されていなければ、対立の回避、割り込み、口癖、傾聴、ファシリテーションなどから尋ねます。
- 傾向を見つけ、それぞれについて所見、頻度、タイムスタンプ付きの引用、重要な理由、より良い言い方を示します。
- 強み、伸ばすべき点、発言の統計、次の一歩をまとめ、期間の比較もできます。
向いている場面
自己コーチング、人事評価の準備、部下へのコーチング。
知っておきたいこと
私たちの試用では、作成した短い架空の文字起こし 2 件に対し、ぼかした発言を正確に引用し、同じ人が対立にうまく対処した場面も指摘し、2 回の会議ではサンプルが小さいとも述べました。
指示だけのファイルで、スクリプトはありません。 指定した会議の文字起こしを読み取ります。他の人の発言が含まれるため、分析する権限があるかを確認し、データはローカルに置いてください。フィードバックは短いテキストに対するモデルの解釈で、客観的な評価ではありません。他の人を評価するのに使う場合は慎重にしてください。 元の skill フォルダには単独のライセンスファイルがなく、ライセンスの根拠は、リポジトリ全体を Apache-2.0 とする元リポジトリの README の記載です(個別の skill は異なる場合があるとも書かれています)。 パッケージには LICENSE(Apache-2.0 全文)と SOURCE.txt(出典とライセンス根拠のメモ。AIBars が追加)も含まれます。