AgentSkillsCN

podcast-wiki-pipeline

当用户希望运行端到端的流水线，将播客或 YouTube 频道的音频源导入文字稿、分块、本体论与维基框架中时使用。当您需要以清单驱动的方式批量处理音视频内容时，此技能同样适用。若需设计维基本体论，请参考 wiki-ontology-design。

SKILL.md
--- frontmatter
name: podcast-wiki-pipeline
version: 1.0.0
description: "When the user wants to run an end-to-end pipeline to ingest podcast or YouTube feeds into transcripts, chunks, ontology, and wiki scaffolds. Also use for manifest-driven batch processing of audio/video content. For wiki ontology design, see wiki-ontology-design."

Podcast Wiki Pipeline

Use this skill to run a repeatable pipeline on a podcast or YouTube feed.

Quick start

bash
python3 scripts/run_pipeline.py --feed-url "https://example.com/feed.xml" --slug "my-podcast"

This writes output to wiki-projects/projects/<slug>/.

What this skill includes

  • •A manifest-first pipeline with checkpoints.
  • •Overlap-aware VTT cleanup and timestamp markers.
  • •Standardized transcript frontmatter.
  • •Chunking into 500-1500 word segments.
  • •Ontology proposal with approval gates.
  • •Wiki scaffold generation (domains/frameworks/cases/people/episodes).

Approval gates (built into the scripts)

  1. •Source selection: confirm discovered transcript sources before ingest.
  2. •Transcript quality: inspect a small sample and approve.
  3. •Ontology: approve a high-level domain map before scaffolding.

Data layout (per project)

code
projects/<slug>/
├── config.json
├── data/
│   ├── manifest.jsonl
│   ├── transcripts/
│   │   ├── raw/
│   │   └── polished/
│   ├── vtt/
│   ├── cues/
│   ├── chunks/
│   └── extracted/
└── wiki/
    ├── ontology/
    └── content/
        ├── domains/
        ├── frameworks/
        ├── cases/
        ├── people/
        └── episodes/

Scripts

  • •scripts/run_pipeline.py — Orchestrates the workflow with approval gates.
  • •scripts/discover_sources.py — Parses RSS/YouTube links and builds manifest.
  • •scripts/fetch_transcripts.py — Downloads transcript sources and builds markdown.
  • •scripts/chunk_transcripts.py — Splits transcripts into semantic chunks.
  • •scripts/propose_ontology.py — Generates a high-level ontology proposal.
  • •scripts/generate_wiki_scaffold.py — Builds wiki stub pages from approved ontology.

Audio-only feeds (optional STT)

If transcripts are missing and you want local STT:

bash
python3 scripts/fetch_transcripts.py \
  --project-root projects/<slug> \
  --stt-command "whisper --model small --language en --output_format txt --output_dir {output} {audio}"

Replace the command with your preferred speech-to-text tool.

References (kept inside this skill)

  • •references/field-notes.md
  • •references/frontmatter.md
  • •references/ontology-blur-to-focus.md
  • •references/manifest-schema.md

Related Skills

  • •Upstream: transcript-polisher, multi-source-research, wiki-ontology-design
  • •Enhanced by: semantic-chunking-rules, local-markdown-search