Crier

---
name: crier
description: "Cross-post blog content to social platforms. Crier is MCP-native; the tools handle mechanics while this skill provides judgment context they can't: rewrite guidelines, platform culture, workflow tips."
---

# Crier Cross-Posting

Crier cross-posts blog content to multiple platforms. The blog is the source of truth.

**Division of labor:**
- **Crier MCP tools** handle mechanics: publish, track, query, validate
- **Claude** handles judgment: rewrites, platform voice, user interaction

## Platform Reference

| Platform  | Mode   | Limit  | Short-form? | Notes                          |
|-----------|--------|--------|-------------|--------------------------------|
| devto     | API    | none   | No          | Tags auto-sanitized            |
| hashnode  | API    | none   | No          |                                |
| bluesky   | API    | 300    | Yes         | Needs rewrite                  |
| mastodon  | API    | 500    | Yes         | Needs rewrite                  |
| medium    | import | none   | No          | User imports from canonical URL |
| twitter   | paste  | 280    | Yes         | Copy-paste only                |
| threads   | paste  | 500    | Yes         | Copy-paste only                |
| linkedin  | paste  | 3000   | No          | Copy-paste only                |

## Rewrite Guidelines

When cross-posting to short-form platforms (bluesky, mastodon), write a rewrite
and pass it via `crier_publish(rewrite_content=...)`.

### Rules

1. **Lead with insight, not summary.** Open with the most interesting idea.
2. **No meta-language.** Don't say "New blog post" or "I wrote about."
3. **Crier appends the canonical URL automatically.** Never include it in the rewrite.
4. **Match platform voice:**
   - Bluesky: conversational, personal, like talking to a friend
   - Mastodon: slightly more formal, technical audience
5. **Budget for the appended URL.** Crier appends `\n\n<canonical URL>` to the
   rewrite and then enforces the platform limit on the combined text, so the
   real budget is `limit - (URL length + 2)`: roughly 230 chars on Bluesky
   (limit 300) and 430 on Mastodon (limit 500) for typical metafunctor URLs.
   Over-budget rewrites fail at the confirmed publish step, not at preview.

### Anti-patterns

- "Check out my latest post on X" (meta)
- "In this article, I explore..." (academic tone for social)
- Including the URL in the rewrite text (crier adds it)
- Generic summaries that could describe any article

### Good example

Article: "pagevault: Hiding an Encryption Platform Inside HTML"
Bluesky rewrite: "What if a single HTML file could encrypt anything you drop into it? No backend, no JS libraries. Just the browser's Web Crypto API and some careful engineering."

## Workflow

### Quick publish (single file)

1. `crier_check(file_path)` to validate
2. `crier_publish(file_path, platform)` with no token: returns a preview plus a `confirmation_token`
3. `crier_publish(file_path, platform, confirmation_token=...)` to execute
4. For short-form: read the article, write the rewrite, pass it as `rewrite_content` on BOTH calls

(`dry_run=True` is a separate no-side-effects preview; it does NOT return a
confirmation token, so it cannot start the two-step flow.)

### Failures: never republish a post that went live

A failed `crier_publish` comes in two shapes, and they need opposite handling:

- **`{"error": "Publish failed: ..."}`**: nothing reached the platform. Safe to
  retry after fixing the cause (auth, length, rate limit).
- **`{"error": "Published remotely but registry recording failed: ...",
  "remote_published": true, "article_id": ..., "url": ...}`**: the post IS live;
  only the local registry write failed. **Do not call `crier_publish` again.**
  The step-1 preview's `already_published` flag reads that same registry, so it
  will not catch the duplicate, and short-form platforms like Bluesky have no
  update path. Instead, repair the record with `crier_record(title, platform,
  platform_id=<article_id>, url=<url>, source_file=<file>)` using the values from
  the error. If that also fails, tell the user to run `crier audit --retry` (CLI),
  which replays crier's saved recovery receipt without republishing.

For retrying old failures in bulk, prefer `crier audit --retry --dry-run` then
`crier audit --retry` over `crier_failures` plus `crier_publish`: the CLI path
checks recovery receipts first, the MCP publish path does not.

### Audit and bulk publish

1. `crier_search(since="2w")` to find recent content
2. `crier_missing(platforms=["devto","hashnode","bluesky","mastodon"])` for gaps
3. For bulk work, dispatch the **cross-poster** agent
4. For analysis, dispatch the **auditor** agent

Agents cannot ask the user anything (Claude Code removes AskUserQuestion from subagents), so the asking happens here, in the main session: get the user's approval for the exact scope BEFORE dispatching the cross-poster, and pass that scope in the launch prompt. When either agent returns, present its `## Decisions Needed` items (cross-poster) or recommendations menu (auditor) with AskUserQuestion, then act on the answers or re-dispatch with the decided scope.

### Important rules

- DevTo sanitizes tags: no hyphens, max 4 tags, lowercase
- Medium is import-only: `crier_publish` refuses import-mode platforms with an
  error; give the user the canonical URL to import on Medium, or use the CLI
  (`crier publish <file> --to medium`)
- Twitter/LinkedIn are paste-only: content goes to clipboard
- To skip social platforms: `crier audit --long-form` (CLI), or read the
  `crier://platforms` MCP resource and filter on `is_short_form`