transcript-analysis

Two self-contained Claude skills for working with meeting recordings:

• preprocess-transcript — pre-processes voice-to-text output (VTT / SRT / TXT), flagging recognition errors by cross-referencing reference documents, and produces a corrections document.
• enrich-meeting-notes — combines raw notes, a transcript, and the transcript analysis into a structured compte-rendu with a table of contents, decisions, action items, diagrams, and highlights.

The two compose into a pipeline: preprocess-transcript writes a corrections document that enrich-meeting-notes then consumes.

Pre-release (0.1.0). These are prompt-based skills — the behaviour lives in each SKILL.md; there is no Python implementation yet. See ROADMAP.md for what a deeper implementation would add.

Layout

skills/
  preprocess-transcript/SKILL.md   # self-contained skill
  enrich-meeting-notes/SKILL.md    # self-contained skill
transcript-analysis.toml           # default folders + output language
.claude-plugin/                    # plugin + marketplace manifests

Each SKILL.md is self-contained: its body is the full prompt. Non-Claude tools can use it directly — copy everything below the YAML frontmatter.

Install

Claude Code plugin (recommended)

/plugin marketplace add teragone-factory/transcript-analysis
/plugin install transcript-analysis@teragone-factory

npx skills (other agents — Codex, Cursor, OpenCode, …)

The Vercel skills CLI discovers the skills from the same repo:

npx skills add teragone-factory/transcript-analysis

Manual copy

ln -s "$PWD"/skills/* ~/.claude/skills/

Skills trigger on phrases like "clean up this transcript" or "enrich these meeting notes from the transcript".

Configuration

Default folder layout and output language live in transcript-analysis.toml:

Key	Default	Purpose
paths.meetings_dir	meetings	Raw notes + enriched output
paths.transcripts_dir	meetings/transcripts	Voice-to-text transcripts
paths.analysis_dir	analysis	Corrections documents
naming.stem	{date}-{slug}	Shared stem linking the files
output.language	fr	Language of the compte-rendu

A meeting's transcript, analysis, and notes share one {date}-{slug} stem (e.g. 2026-03-25-design-review). That stem is the contract that lets the two skills find each other's files.

First run. The file ships as the canonical template. When a skill runs in a target repo that has no transcript-analysis.toml, it proposes the defaults above — guessing output.language from the repo's CLAUDE.md — notifies you, and asks you to confirm or adjust before doing anything. Once confirmed it writes the file into the target repo; later runs read it silently.

Licence

Copyright 2026 Teragone Factory. Licensed under Apache-2.0; see NOTICE.