PodStack

One repo, one install, entire podcast content workflow. It works with Claude Code, OpenAI Codex, Cursor, opencode, or any AI tool that reads markdown.

PodStack turns your AI coding tool into a full podcast content team using structured skill files. It gives you an Episode Architect, Content Analyst, Title Writer, Copywriter, Art Director, Brand Guardian, Producer, Launch Manager, and Performance Analyst, all driven from your terminal or editor.

Built for podcasters who use podcli but works standalone with any project.

Inspired by gstack by Garry Tan.

What You Get

Command	Role	What It Does
`/plan-episode`	Episode Architect	Designs questions, story arc, moment map BEFORE recording
`/process-transcript`	Content Analyst	Ingests transcript → extracts best moments → scores → categorizes
`/generate-titles`	Title Writer	Generates 8 title options with full verification checklist
`/generate-descriptions`	Copywriter	Creates descriptions + hashtags + SEO keywords
`/plan-thumbnails`	Art Director	Plans thumbnail text + layout briefs for your designer
`/review-content`	Brand Guardian	Fix-First review: auto-fixes + parallel specialist dispatch
`/produce-shorts`	Producer	Full pipeline: transcript → publish-ready package
`/publish-checklist`	Launch Manager	Pre/post-publish optimization checklist
`/retro-episode`	Analyst	Episode performance review + appends patterns to `knowledge/13-learnings.md`

Host Compatibility

PodStack ships one source-of-truth in commands/ and installs to the right location for whichever tool you use:

Host	Install location	Primary instruction doc
Claude Code (default)	`.claude/commands/*.md`	`CLAUDE.md`
OpenAI Codex	`.codex/prompts/*.md`	`AGENTS.md`
Cursor	`.cursor/rules/*.mdc`	`AGENTS.md`
opencode	`.opencode/commands/*.md`	`AGENTS.md`
Generic (anything else)	`commands/*.md`	`AGENTS.md`

setup.sh auto-detects your tool based on existing .claude/, .codex/, .cursor/, or .opencode/ folders. Override with --host.

Install

Quick install (auto-detects your tool)

Clone PodStack into your project as .podstack, then run its installer from the project root:

git clone https://github.com/nmbrthirteen/podstack.git .podstack
.podstack/setup.sh

The installer picks the host from existing .claude/, .codex/, .cursor/, or .opencode/ folders and defaults to Claude Code. In a podcli project (a .podcli/ folder exists) it puts the knowledge base in .podcli/knowledge/.

Running setup.sh from inside a clone named .podstack installs into the parent folder. Any other clone location needs the project path: setup.sh <project-dir>.

Explicit host

# For Claude Code
.podstack/setup.sh --host claude-code

# For OpenAI Codex
.podstack/setup.sh --host codex

# For Cursor (converts .md → .mdc rules)
.podstack/setup.sh --host cursor

# For opencode
.podstack/setup.sh --host opencode

# For any other tool (copies to project-root commands/ + AGENTS.md)
.podstack/setup.sh --host generic

Update

git -C .podstack pull
.podstack/setup.sh --update

--update overwrites the commands PodStack installed. It records them in a .podstack-manifest file in the commands folder. If you edited an installed command, --update saves your version as <file>.bak first. Without --update, existing files are skipped.

Shared clone for several projects

git clone https://github.com/nmbrthirteen/podstack.git ~/.podstack
cd ~/my-podcast && ~/.podstack/setup.sh

Each project gets its own copy of the commands. Run setup.sh once per project.

Usage by Host

Claude Code

/plan-episode
/produce-shorts
/retro-episode

Commands appear in /help. Claude reads CLAUDE.md automatically.

OpenAI Codex

Codex reads AGENTS.md at project root on startup. Invoke skills by reference:

Run the produce-shorts playbook on transcripts/ep42.txt
Use the generate-titles skill for this clip: [paste]

Codex loads the referenced file from .codex/prompts/ and follows it.

Cursor

Rules live in .cursor/rules/*.mdc. Invoke in chat:

Use the produce-shorts rule on this transcript
Apply the review-content rule to the content package at episodes/ep42.md

Cursor matches rules by description.

opencode

Commands work like Claude Code slash commands:

/plan-episode
/produce-shorts

Generic tool

Point your AI at AGENTS.md and reference skills by path:

Follow the instructions in commands/produce-shorts.md using this transcript: [paste]

Quick Start

1. Set up your brand

Edit the knowledge base files in knowledge/ (or .podcli/knowledge/ if using podcli):

knowledge/
├── 00-master-instructions.md    ← Your show's AI operating system
├── 01-brand-identity.md         ← Who you are, positioning, hosts
├── 02-voice-and-tone.md         ← How you sound, banned words
├── 03-episodes-database.md      ← Track episodes and existing shorts
├── 04-shorts-creation-guide.md  ← Moment selection criteria
├── 05-title-formulas.md         ← Your title patterns and rules
├── 06-descriptions-template.md  ← Description templates + hashtags
├── 07-thumbnail-guide.md        ← Visual specs for your brand
├── 08-topics-themes.md          ← Your content categories
├── 09-content-workflow.md       ← End-to-end workflow
├── 10-internal-processing.md    ← Internal rules for your AI
├── 11-inspiration-channels.md   ← Reference channels you admire
├── 12-quick-reference.md        ← Copy-paste resources
└── 13-learnings.md              ← Cross-episode learnings (auto-updated by /retro-episode)

Each file ships as a starter template with examples. Fill them in with your show's specifics.

2. Process your first episode

/produce-shorts

Paste your transcript or point to a file. PodStack runs the full pipeline and delivers a publish-ready content package.

3. Or use individual commands

/process-transcript     ← Just extract moments from a transcript
/generate-titles        ← Just get title options for a specific clip
/review-content         ← Just review existing content for quality

The Pipeline (Sprint Workflow)

Plan → Process → Write → Review → Publish → Retro

/plan-episode  →  [record]  →  /process-transcript  →  /generate-titles  →  /generate-descriptions
                                                                                    ↓
             /retro-episode  ←  [published]  ←  /publish-checklist  ←  /review-content  ←  /plan-thumbnails

Or run everything at once: /produce-shorts

After publishing: /retro-episode also appends cross-episode learnings so future runs get smarter.

How It Works

PodStack skills are plain markdown files with YAML frontmatter. Each one shifts your AI into a specialized role. Every skill includes:

Frontmatter: description, allowed-tools, argument-hint, triggers. Claude Code reads description, so put trigger phrases there. triggers is guidance for people and for tools that read it; Claude Code ignores that key
A clear role: what the agent is acting as (analyst, writer, reviewer)
An execution protocol: step-by-step instructions
Quality gates: checks that must pass before output
Output format: structured, consistent deliverables
Completion Protocol: every skill returns DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_INPUT with evidence
Three-Strike Rule: orchestrators abort after 3 consecutive phase failures instead of looping
Reference file pointers: links to your knowledge base for detailed specs

The knowledge base files are your show's "brand brain". They contain your voice, style, title formulas, banned words, and everything needed to produce content that sounds like you, not generic AI.

/review-content also uses Fix-First with parallel specialist dispatch. It auto-applies mechanical fixes (banned words, hashtag dedup, casing) and batches human-decision items into one question.

Customization

Add your own skills

Create any .md file in commands/ (or your host's command folder) with the frontmatter format:

---
description: What the skill does. Use when the user says phrase 1 or phrase 2.
allowed-tools: Read, Write
argument-hint: [what the user types]
triggers:
  - natural language phrase 1
  - natural language phrase 2
---

# /my-custom-skill: My Role

> Instructions for the agent...

## Execution Protocol
...

## Completion
Return DONE / DONE_WITH_CONCERNS / BLOCKED / NEEDS_INPUT per CLAUDE.md or AGENTS.md.

If you add the file to the PodStack clone's commands/, run .podstack/setup.sh to install it for your host. A file placed directly in your host's command folder works as is. --update never touches files PodStack did not install.

Modify existing skills

All skills are plain markdown. Edit them to match your workflow. Common customizations:

Change the number of title options (default: 8)
Add your own quality gates
Add new specialists to /review-content
Modify output formats
Add new content types beyond the default five

podcli integration

When the podcli MCP server is connected, /process-transcript and /produce-shorts use its tools. Without it they work from text files.

transcribe_podcast: transcribes an audio or video file when you have no transcript
suggest_clips: adds AI-suggested moments to the candidate pool
clip_history: checks which clips already exist, to avoid duplicates
knowledge_base: reads your brand files when they are not on disk
create_clip: renders approved moments with captions at the end of /produce-shorts

Project Structure

podstack/
├── README.md                    ← You are here
├── CLAUDE.md                    ← Claude Code-specific primary doc
├── AGENTS.md                    ← Host-agnostic primary doc (Codex, Cursor, opencode, generic)
├── ETHOS.md                     ← Content philosophy (the why behind every rule)
├── CHANGELOG.md                 ← Release history
├── VERSION                      ← Current version
├── LICENSE
├── setup.sh                     ← Host-detecting installer (--host, --update)
├── scripts/
│   ├── sync-to-podcli.sh        ← Sync commands and templates into a podcli project
│   └── manifest.sh              ← Shared helpers: install manifest and checksums
├── commands/                    ← Skill files (with YAML frontmatter), source of truth
│   ├── plan-episode.md
│   ├── process-transcript.md
│   ├── generate-titles.md
│   ├── generate-descriptions.md
│   ├── plan-thumbnails.md
│   ├── review-content.md
│   ├── produce-shorts.md
│   ├── publish-checklist.md
│   └── retro-episode.md
└── templates/                   ← Starter knowledge base files
    ├── 00-master-instructions.md
    ├── 01-brand-identity.md
    ├── 02-voice-and-tone.md
    ├── 03-episodes-database.md
    ├── 04-shorts-creation-guide.md
    ├── 05-title-formulas.md
    ├── 06-descriptions-template.md
    ├── 07-thumbnail-guide.md
    ├── 08-topics-themes.md
    ├── 09-content-workflow.md
    ├── 10-internal-processing.md
    ├── 11-inspiration-channels.md
    ├── 12-quick-reference.md
    └── 13-learnings.md

Requirements

One of: Claude Code, OpenAI Codex, Cursor, opencode, or any AI tool that reads markdown
Optional: podcli for video processing

License

MIT

Navigation Menu

Repository navigation

README.md

README.md

PodStack

What You Get

Host Compatibility

Install

Quick install (auto-detects your tool)

Explicit host

Update

Shared clone for several projects

Usage by Host

Claude Code

OpenAI Codex

Cursor

opencode

Generic tool

Quick Start

1. Set up your brand

2. Process your first episode

3. Or use individual commands

The Pipeline (Sprint Workflow)

How It Works

Customization

Add your own skills

Modify existing skills

podcli integration

Project Structure

Requirements

License

FilesExpand file tree

README.md

Latest commit

History

README.md

File metadata and controls

PodStack

What You Get

Host Compatibility

Install

Quick install (auto-detects your tool)

Explicit host

Update

Shared clone for several projects

Usage by Host

Claude Code

OpenAI Codex

Cursor

opencode

Generic tool

Quick Start

1. Set up your brand

2. Process your first episode

3. Or use individual commands

The Pipeline (Sprint Workflow)

How It Works

Customization

Add your own skills

Modify existing skills

podcli integration

Project Structure

Requirements

License