Transcript Processor Skill

A precise workflow tool designed to clean, structure, and optionally analyze raw interview or podcast transcripts into high-fidelity Q&A digests without losing original context.

Workflow

1. State the Plan: Infer mode and target language, present the execution plan, and wait for explicit user confirmation.
2. Run Parser: Execute python3 scripts/parse_transcript.py <file> to generate a timestamp skeleton to prevent missing any segments. Skipped in upgrade mode.
3. Clean & Format: Clean filler words/false starts while strictly preserving the speaker's original words (no paraphrasing/summarizing).
4. Translate (if applicable): Translate Q/A into target language after cleaning. Skipped if output language matches transcript language.
5. Analysis (Digest Mode Only): Append hard-hitting commentary for each exchange and a final meta-analysis section.

Modes

• Clean Mode (Default): Outputs strict Q (with timestamp) and A blocks only. No external summary or commentary.
• Digest Mode: Adds deep analysis layer (Commentary) identifying avoided tensions, verifiable claims, and a concluding Meta-analysis of recurring patterns.
• Upgrade Mode: Takes an existing clean .md file and adds Commentary and Meta-analysis directly. Skips parser and clean steps entirely; Q and A text is read-only.

File Output

Saves results into [interviewee]_[source]_[mode].md and runs scripts/check_digest.py before completion.