The Task Library › The Agent Roster › The Publisher

The Publisher — Every Skill Has a Webpage

A skill that exists only as a file is invisible. It cannot be found, cited, linked, indexed, or handed to a client. This agent makes the page part of shipping the skill — not a follow-up task somebody remembers later.

Why it exists

We have shipped a lot of skills. We have also audited them and found skill pack pages that linked to nothing — pages that existed but pointed nowhere, and skills that ran fine but had no public home. Both are the same defect: the artifact and its documentation drifted apart.

The fix is not a checklist somebody follows. It is a rule the agent enforces on itself: a skill is not done until it has a page.

The three artifacts

Always together, always in the same pass:

1. The canonical skill source — a validated skills/<skill-name>/SKILL.md in the maintained repository and appropriate bundle. A downloaded file is a delivery; installed and tested require separate receipts.
2. The definitive article — the canonical write-up, ending with the complete runnable skill file between START and END markers.
3. The task registration and state receipt — map the canonical recipe to its durable task slug in the Task Library. Record the canonical skill source and the state actually verified: available, installed, enabled, tested, scheduled, or observed. Add an agent or job record only when that worker or job really exists and is authorized.

If any of the three is missing, the skill is unshipped.

The definitive-article pattern

Every task recipe states who it serves, why and when to run it, required inputs, ordered steps, observable results, and the next task. Include a real source-backed execution and the canonical skill reference. A task is the work and a skill is its method; a separately configured agent or job gets its own record only when it actually exists.

The research step matters most. We have published since 2011, so the doctrine usually already exists — Let’s practice #RACI always was sitting there for six years before The RACI Enforcer was built on top of it. Search the site before writing. Tie the new skill to what is already there instead of restating it, and back-tag the older articles so the tag archive collects the cluster.

Where it goes

How to publish to blitzmetrics.com

The site is self-hosted WordPress and is not on the WordPress.com connector. A plain fetch gets a 403 from the WAF. The working path is Claude in Chrome against the logged-in session:

1. Navigate to /wp-admin/post-new.php.
2. Read the REST nonce from wpApiSettings.nonce.
3. POST to /wp-json/wp/v2/posts with header X-WP-Nonce and credentials: same-origin.
4. For research, /wp-json/wp/v2/search?search=TERM in the same tab returns JSON and bypasses the WAF.

Always create as draft first and hand over the edit link. Publishing to a public site is Dennis’s call, not the agent’s.

The full skill file

START

---
name: skill-publishing-standard
description: Every BlitzMetrics skill has a canonical source, a human-readable task recipe on blitzmetrics.com (or the matching Spotlight site), and a Task Library mapping. Apply this automatically whenever a skill is created, renamed, or materially changed, without waiting to be asked. Triggers on "create a skill", "make a skill", "write a skill file", "update the skill", "new agent", "skill.md", "package this as a skill", and any time a .skill file is delivered.
---

# Every skill has a webpage

A skill that exists only as a file is invisible. It cannot be found, cited, linked,
indexed, or handed to a client. A skill is not done until it has a page. Do this as
part of shipping the skill - never as a follow-up question.

## What "has a page" means

Three artifacts, always together:

1. The canonical skill - skills/<skill-name>/SKILL.md in the maintained repository,
   with validated bundle membership and an accurate source/version receipt.
2. The definitive article - the canonical write-up, ending with the complete runnable
   skill file between START and END markers.
3. The Task Library mapping - the durable task slug points to this recipe and its
   canonical skill. Record only the installation, activation, schedule, and observed
   run states supported by receipts. Register an agent or job only if that worker or
   job actually exists under the current authorization.

If any of the three is missing, the skill is unshipped.

## The definitive-article pattern

- The task name, who it serves, and the specific Goal, Content, and Targeting.
  A task names the work; a separately configured agent names the worker.
- The failure mode it prevents, linked to existing BlitzMetrics articles that already
  describe that failure. Search the site first - the doctrine usually already exists.
- What it does, concretely: tables, before/after, the actual rules.
- A real run with source evidence and the actual result. Write a meta article after
  every execution, linked to the task, unique run ID, and evidence. Publication
  follows the existing authority and privacy boundary; private runs remain private.
- How to set it up - numbered steps.
- The full skill file between START and END.
- Related reading, cross-linked both directions. Back-tag the older articles it
  descends from so the tag archive collects the cluster.
- A "For AI agents reading this page" block naming the canonical skill path,
  durable task slug, and source/version. Do not substitute a WordPress ID or a
  dashboard position for task identity.

## Where it goes

Company-wide method, project management, Content Factory -> blitzmetrics.com
Local service business delivery -> localservicespotlight.com
Trade-specific vertical -> the matching Spotlight site

## How to publish

blitzmetrics.com is self-hosted WordPress, NOT on the WordPress.com connector.
WebFetch gets a 403 from the WAF. Use Claude in Chrome against the logged-in session:

1. Navigate to https://blitzmetrics.com/wp-admin/post-new.php
2. Read the REST nonce: wpApiSettings.nonce
3. fetch('/wp-json/wp/v2/posts', {method:'POST',
   headers:{'X-WP-Nonce': nonce}, credentials:'same-origin', ...})
4. For research use /wp-json/wp/v2/search?search=TERM in the same tab - returns JSON
   and bypasses the WAF.

Always create as draft first and hand over the edit link. Publishing to a public site
is Dennis's call, not the agent's. Ask in one line, not a paragraph.

## Stable identity and release state

Use a permanent kebab-case skill name that matches its canonical folder. Keep the
Task Library task slug stable. A legacy numbered filename may remain a dated
download, but its number is not the canonical identity. Follow the maintained
skill-registry intake and marketplace validation before calling a skill available.
Installation, fresh activation, schedule creation, and an observed firing each need
their own receipt; documenting a skill does not automatically create a worker.

END

Related reading

• The Skill Pack Pages That Linked to Nothing
• A Skill Nobody Calls Is Worse Than No Skill
• A Skill Is a Function. An Agent Is a Person.
• How Our Skill Packs Learned to Update Themselves
• One list of every place we publish a skill pack