meeting-analysis

Convert meeting recordings, transcripts, and notes into citation-bearing analysis documents.

Updated Sep 10, 2026
One-click install
npx skills add https://github.com/sergeiwallace/skills --skill meeting-analysis-sergeiwallace
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: meeting-analysis
Source: https://github.com/sergeiwallace/skills/tree/main/skills/meeting-analysis
Command: npx skills add https://github.com/sergeiwallace/skills --skill meeting-analysis-sergeiwallace

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires faster-whisper, pillow, imagehash, numpy, typer, pydantic-settings, skill-cli-foundation, and includes scripts (resource) components.

What problem does it solve? Turning raw meeting artifacts (recordings, vendor transcripts, notes exports) into an auditable written analysis is error-prone: vendor transcripts mangle product names, recordings can be truncated while claiming full duration, and pixel-based frame selection often fails on continuous screen shares. This Skill provides a deterministic workflow that cross-checks every claim against its sources. ## Core Features & Use Cases - Recording integrity verification: Detects recordings whose containers declare more duration than their payload contains, and reports the usable bound before any analysis sizes work from it. - Independent ASR cross-check: Runs a local faster-whisper pass and diffs it against the vendor transcript to recover mangled product names and terminology, with side-by-side probing to settle mishearings. - Deterministic planning: A resolver script picks the correct doc kind (recording analysis vs. standup update), derives conventional paths, classifies sources, and refuses to overwrite committed docs. - Use Case: You receive a zip with a standup recording, a vendor VTT transcript, and a notes PDF. The Skill extracts the sources, verifies the recording, runs ASR, corrects transcript terminology, reads screen-share frames, and produces a template-conformant analysis doc with timestamps and provenance. ## Quick Start Ask the assistant to analyze the meeting by running /meeting-analysis on your zip or directory of meeting sources with your name and the meeting date.

Frequently Asked Questions about meeting-analysis

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I analyze a meeting recording and transcript into a written summary?

Provide a zip or directory containing the recording, vendor VTT transcript, and notes PDF. The workflow verifies recording integrity, runs a local faster-whisper ASR pass, diffs it against the vendor transcript, reads screen-share frames, and fills a template-conformant analysis doc.

How do I fix wrong product names in a vendor meeting transcript?

Run a local ASR pass with faster-whisper, then use transcript_diff.py to find terms the ASR heard but the vendor never wrote. Feed those candidates into probe_terms.py to see both sources side by side and settle each mishearing with evidence.

Why does my meeting recording show a longer duration than it actually contains?

Some containers declare more duration than their decodable payload holds, and ffprobe's duration field reports the declared figure. recording_integrity.py compares declared duration against the last decodable packet per stream and reports the usable bound to cap ffmpeg extractions.

Does frame deduplication work for screen-share recordings of console work?

Often not. Continuous console or browser work produces a unimodal perceptual-hash distance distribution with no clean change signal. Run phash_distribution.py to check; if unimodal, anchor frame selection on the transcript's screen-share cues instead of pixel thresholds.

What dependencies are required to run the meeting analysis scripts?

ffprobe for integrity checks, ffmpeg for audio extraction and frame grids, pdftotext for notes PDFs, and Python packages faster-whisper, pillow, imagehash, and numpy. The check_deps.sh preflight verifies them by importing, and installation is opt-in via MEETING_ANALYSIS_AUTO_INSTALL=1.