Open Context Forge Article Publisher

Language: English · 简体中文

Take a reviewed article from a writing project to a channel-specific publication package.

Open Context Forge Article Publisher is a file-based Agent Skill for the article-to-channel stage of Open Context Forge. It selects an existing article, checks for publishing blockers, copies local images, and prepares output for a chosen channel. It does not draft the article or change the knowledge vault. The operational SKILL.md and channel references are currently written in Chinese.

What you need

• An agent that can read the whole Skill directory and access the finished article and its assets.
• Node.js 20 or newer. The publishing script uses only Node.js built-ins.
• A reviewed Markdown or MDX article. Writing projects may contain article.md, published.md, or final.md.
• Write access to the selected export directory. Writing directly to the apframework blog also requires access to that repository.

Use it with an agent

Make this entire repository available in your agent's Skill search path, keeping SKILL.md, references/, and scripts/ together. You can also ask an agent to read SKILL.md by its absolute path without registering the Skill by name.

For example:

Use $open-context-forge-article-publisher to prepare /absolute/path/to/writing-project for apframework. Put the output in /absolute/path/to/review-package, inspect the article and images, and report what is still needed before it goes live.

When a project directory is supplied, the script looks for article.md, then published.md, then final.md. You can name an article file explicitly. If several projects could match a request, the agent asks you to select one instead of guessing from modification time. A legacy vault archive is used only when explicitly named.

Channels and delivery status

Channel	Current output	What it means
apframework	Contentlayer MDX and images in the blog layout	An export is prepared; writing to the blog repository is written-to-repository. Build, deployment, and public URL checks are separate.
wechat	Markdown and local assets for a WeChat Official Account editor	An import package is prepared. It is not an account draft or a sent post.
generic-md	Portable Markdown and relative image paths	A reusable package is prepared; the destination platform still needs its own publishing step.

Only a checked public URL establishes that an article is live. Channel-specific procedures are in references/apframework.md, references/wechat.md, and references/generic-md.md.

An existing apframework article can require a date, filename, image directory, and reference migration. Follow the apframework update procedure before changing files; --replace does not perform that migration. A blog write is written-to-repository; a running build, failed build, successful build, and verified live page are reported separately.

Run the script

Run from this repository's root, or use an absolute path to scripts/publish.mjs:

# Prepare a package for review. The source article is not modified.
node scripts/publish.mjs --project /absolute/path/to/writing-project \
  --channel apframework --output /absolute/path/to/review-package --json

# Prepare an editor import package.
node scripts/publish.mjs --article /absolute/path/to/article.md \
  --channel wechat --output /absolute/path/to/wechat-package --json

# Write to a checked-out blog repository when that is the requested delivery.
node scripts/publish.mjs --project /absolute/path/to/writing-project \
  --channel apframework --blog-root /absolute/path/to/blog \
  --slug english-slug --json

# Read-only check of a written blog article's local image references and date.
node scripts/check-apframework.mjs --blog-root /absolute/path/to/blog \
  --article /absolute/path/to/blog/data/blog/essay/YYYY-MM-DD-english-slug.mdx --json

With --project and no destination option, packages go to <project>/exports/<channel>/. With --article, supply --output or, for apframework, --blog-root. Use --title, --summary, or --date YYYY-MM-DD when the finished article lacks public metadata. Local assets must resolve inside the article project; --asset-root /absolute/allowed/root explicitly allows a wider source tree. Existing outputs are not replaced unless --replace is given.

The script rejects draft status, unresolved wiki links, unfinished markers, common local absolute paths, missing images, and conflicting asset names. A prepared package includes publication.json with a source hash and relative output paths. A direct blog write returns its report on the command line without adding a local-path manifest to the blog repository. The agent still reviews the generated article and performs the channel's build or editor preview.

The read-only apframework checker verifies literal Markdown/MDX image, image-link, and frontmatter images: paths against the blog's public/ directory. It also catches a filename/frontmatter date mismatch and references to the previous date directory for the same slug. For a date and slug migration, pass --previous-id YYYY-MM-DD-old-slug to check the old image directory explicitly. It cannot validate dynamic JSX, image quality, captions, alt text, or licensing.

Project files

• SKILL.md — agent workflow, source selection, and publishing boundary.
• scripts/publish.mjs — deterministic channel export CLI.
• scripts/check-apframework.mjs — read-only blog image-path and date check.
• references/ — channel-specific preparation and verification.
• docs/design.md — feature design and delivery levels.
• docs/implementation-plan.md — initial implementation sequence.
• docs/public-release-readiness.md — checks before making the repository public.
• docs/publishing-feedback.md — adopted, deferred, and rejected publication feedback.
• tests/publish.test.mjs — CLI behavior tests.

Run npm test to check the script. The project is licensed under MIT.