COPE — Create Once, Post Everywhere

Turn one idea into a complete, platform-ready content package.

COPE is an open-source, local-first content repurposing agent. Give it rough notes, a Markdown draft, or a URL; it builds one canonical brief, then creates X posts, LinkedIn posts, blog drafts, social images, and swipeable carousel PDFs from the same source.

pnpm cope run examples/post.md --all

One command produces a reviewable folder of text, images, metadata, and a run summary. COPE never publishes automatically.

Why COPE?

• One source, many formats. Create an X post, thread, LinkedIn post, blog draft, OG image, quote card, and carousel without rebuilding the idea each time.
• Fast local visuals. Information cards render locally with no image-generation or vision calls; use generated artwork only when it adds value.
• Complete carousel copy. Every slide gets a short 5–6 word title and a complete supporting description. Text that cannot fit fails instead of being clipped.
• Bring your preferred model. Use Claude Code, Codex, Anthropic, OpenAI, or a trusted OpenAI-compatible endpoint through configuration.
• Review before publishing. Outputs are files, validation failures are explicit, and every run is isolated so an older result is never silently reused.

Agent or skill?

Describe COPE as an agent. It owns a multi-step workflow: ingesting source material, building a brief, routing work to models, adapting formats, validating results, rendering visuals, and writing artifacts. A Claude or Codex skill is usually an instruction bundle that runs inside its host. COPE is a standalone CLI agent that can use either host as a provider, so a future skill would be a thin way to invoke COPE rather than the product itself.

What it creates

Output	Format ID	Result
X post	x_post	Concise single post
X thread	x_thread	Structured multi-post thread
LinkedIn post	linkedin_post	Professional long-form social post
Blog draft	blog_md	Markdown article
OG image	og_image	Share-ready PNG
Quote card	quote_card	Branded quote PNG
Carousel	carousel	PNG slides, metadata, and a PDF

Start with selected formats when you want a faster first run:

pnpm cope run examples/post.md -f x_post,linkedin_post,carousel

Installation

COPE is currently clone-and-run. It requires Git, Node.js >=20.9, and pnpm 10.11.1:

git clone https://github.com/sirius93/cope.git
cd cope
corepack enable
corepack prepare [email protected] --activate
pnpm install --frozen-lockfile

Then install and authenticate the provider needed by your workflow.

Primary workflow	Local cards and carousels	Generated artwork
Claude Code (default)	Claude writes the copy; COPE renders it locally. Codex is not needed.	Claude keeps the full text workflow; Codex handles only image generation and vision-based alt text.
Codex	Codex writes the copy; COPE renders it locally.	Codex handles the complete workflow.

The built-in card renderer is the default. It creates the carousel PNGs and PDF from predefined layouts, so neither agent needs to generate an image unless you explicitly select --image-mode generated.

Claude Code: default local-card workflow

The default configuration uses Claude Code for the brief, text formats, and carousel copy. Install it using an official Claude Code method:

# macOS, Linux, or WSL
curl -fsSL https://claude.ai/install.sh | bash

# macOS with Homebrew
brew install --cask claude-code

Windows users can use the PowerShell or WinGet commands in the linked setup guide.

Run claude once to sign in, then verify and run COPE:

claude --version
pnpm cope run examples/post.md -f x_post,linkedin_post
pnpm cope run examples/post.md --all --image-mode cards

This path does not require an API key or Codex. Claude Code's account limits, billing, and service terms still apply.

Codex: image handoff or complete workflow

When Claude is the primary agent, the generated-artwork route hands only image generation and vision-based alt text to Codex. Install it using the official Codex CLI installer:

# macOS or Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

Windows users should follow the Windows instructions in the linked Codex guide.

Run codex once and choose a sign-in method, then verify it:

codex --version

With the default configuration, keep Claude installed and add Codex for generated images:

pnpm cope run examples/post.md --all --image-mode generated

To use Codex for the entire pipeline, create cope.config.yaml:

routing:
  brief: codex_cli
  voice_learn: codex_cli
  longform: codex_cli
  midform: codex_cli
  shortform: codex_cli
  edit: codex_cli
  image_prompt: codex_cli
  image_generate: codex_cli
  alt_text: codex_cli

API providers: no agent CLI

Copy .env.example to .env, add ANTHROPIC_API_KEY and/or OPENAI_API_KEY, then route stages to anthropic or openai in cope.config.yaml. For an OpenAI-only setup:

routing:
  brief: openai
  voice_learn: openai
  longform: openai
  midform: openai
  shortform: openai
  edit: openai
  image_prompt: openai
  image_generate: openai
  alt_text: openai

Verify the configured model IDs before running API routes. For all commands, provider details, and Claude/Codex integration notes, see the usage guide.

Fast images

pnpm cope run examples/post.md --all

Images default to local information cards: readable typography, key points, and a carousel PDF rendered from the brief. This mode makes zero image-generation or vision calls. A carousel uses one text call to create complete 5–6 word titles and supporting descriptions; other cards render directly from the brief. Alt-text is derived from the card's actual content. No codex installation is needed.

For AI-generated artwork instead:

pnpm cope run examples/post.md --all --image-mode generated

Generated mode keeps the brief, writing, editing, and image concept with the selected primary agent. By default, only image generation and vision-based alt text use the local codex CLI. Those calls can take many minutes, and tool availability depends on the installed CLI and account. This integration is experimental; the automated tests use substitutes, not live authenticated services. Simpler visual prompts alone do not guarantee faster remote generation.

Visual styles

Choose a style independently of image mode:

pnpm cope styles
pnpm cope run examples/post.md --all --image-style editorial
pnpm cope run examples/post.md -f carousel --image-mode generated --image-style sketch

Style	Look
editorial (default)	Warm paper, serif headlines, restrained magazine composition
minimal	White space, centered sans-serif typography, monochrome geometry
blueprint	Deep navy, fine construction lines, technical typography
sketch	Off-white paper, graphite-like accents, quiet notebook feel

Local cards use the selected palette, typography and layout without extra model calls. Generated mode applies the art direction to both concept and image prompts and uses a matching caption panel. Actual artwork still depends on the image provider. Styles stay consistent within a carousel and are recorded in each image's JSON metadata. There are no automatic COPE banners or watermarks.

Output and validation

Each run gets a fresh out/<slug>-<unique-id>/ directory, containing brief.json, run-summary.json, text drafts, and images/ (PNG, metadata/alt-text JSON, carousel PDF). Previous runs are never reused. Failed text drafts go under invalid/; any failed format sets a nonzero exit code. Other formats continue after a provider failure.

Validation checks JSON types and configured numeric limits, with a bounded repair loop. It does not establish factual accuracy, originality, or suitability for publication. Failed image formats may leave partial files; consult run-summary.json before using them.

Formats: x_post, x_thread, linkedin_post, blog_md, og_image, quote_card, carousel. The og_image card contains a headline; generated mode uses artwork without a text overlay.

Configuration and providers

config/defaults.yaml contains defaults. A local cope.config.yaml merges sections and individual provider entries, so a small override is enough:

images:
  mode: cards # or generated
  style: editorial # minimal | blueprint | sketch
limits:
  maxImagesPerRun: 10
voice:
  file: persona/voice.md # gitignored by default — it's derived from your own writing, not project code

Brand images with a logo, colors, website, and social handle by copying persona/persona.yaml.example to persona/persona.yaml — see docs/USAGE.md for the fields. It's optional and gitignored, same as persona/voice.md.

Provider	Authentication	Use
claude_cli	Existing local CLI login	Text (default)
codex_cli	Existing local CLI login	Text, experimental generated artwork and vision
anthropic	ANTHROPIC_API_KEY	Text API
openai	OPENAI_API_KEY	OpenAI-compatible text, image and vision APIs

Copy .env.example to .env for API credentials and override the relevant routing entries. Verify model IDs with your provider before using the API routes. Custom baseUrl values receive the configured credentials and submitted content; only use trusted endpoints. See usage and security.

maxImagesPerRun bounds the total image count reserved across a run, including failed formats. Image generation and CLI subprocesses are not automatically retried: a failed request may still be billable. Text HTTP requests retry only transient HTTP statuses. maxRepairPasses is limited to 0–10. maxTokensPerRun is unsupported and rejected; configure spending/quota limits with your provider. This is not a dollar-budget guarantee.

Development

pnpm typecheck
pnpm test

Source packages live in packages/{providers,formats,render,core,cli}. Distribution is currently clone-and-run; packages are private and are not prepared for npm publication. See contributing. The original build plan is historical planning material, not a statement of implemented features.

License

COPE's original code and documentation are MIT licensed. Dependencies retain their own licenses; see third-party licensing, especially before redistributing native binaries or containers. The software license does not determine rights to source material or generated output, or replace provider service terms.