$ cat ./HARNESS.md

This blog is also the place I test how I work with agents. A set of subagents, skills, hooks and CI checks drafts, reviews and ships it, under rules that keep the writing mine. This page is the log of that harness: the posts about it, short notes as it changes, and what it is made of today.

## Posts

## Notes

> The first notes on how the harness changes are on their way.

## The harness today

Read from the repository when this page was built: 8 subagents, 9 skills, 4 hook scripts and 15 workflows.

Subagents

autoharness-curator
Periodic consolidation pass — fold narrow agent-created skills into class-level umbrellas.
autoharness-reflector
Distill a finished episode into skill changes aligned with the existing library.
blog-pm
Use this agent for product-oriented improvements to the blog's mechanics, discoverability, and reach — including SEO, structured data, LLM-readability (so ChatGPT/Perplexity/Claude/etc. cite the blog accurately),…
blog-qa
Use this agent to perform a QA + regression sweep over the blog repository — checking CI health, tracked issues, broken links, RSS/sitemap correctness, accessibility, perf budgets, and other operational hygiene.
editor-in-chief
Use this agent when you need editorial guidance for blog content—from initial brainstorming through final publication.
harness-curator
Use this agent for the monthly review of the agent harness: run the harness audit, read what reviewers corrected on recent routine pull requests, and propose one draft pull request of small, tested fixes to agents,…
kobi-personal-assistant
Use this agent when you need personalized assistance, strategic advice, or information retrieval that requires deep knowledge of Kobi Kadosh's background, preferences, and context.
linkedin-role-enhancer
Use this agent when you need to transform basic job descriptions into compelling, outcome-driven LinkedIn Experience entries that include company context, measurable achievements, and external validation.

Skills

codify
Use when a mistake recurs, a reviewer repeats a correction, or a written rule keeps being ignored, to turn the lesson into a test, hook or check.
content-quality
Use when reviewing a blog post's quality — prose, readability metrics, spelling, inclusive language, Markdown structure, or a TLDR.
field-notes
Use when harvesting, reviewing, writing or publishing a field note (apps/site/src/content/notes/), when the weekly "Field notes inbox" issue arrives, or when Kobi asks to turn a session, a week of work or a link into a…
harness-audit
Use when auditing a repository's agent harness (hooks, skills, subagents, CLAUDE.md, AGENTS.md, Claude workflows, .mcp.json) for failures that make no noise.
learn
Use when the user says /learn or asks to save this lesson.
post-images
Use when a blog post (apps/site/src/content/blog/*.mdx) needs its cover image picked, its share (Open Graph) card checked, or image variants made, and whenever the image bot has posted candidates on a pull request.
post-republish
Use when a blog post or TIL (apps/site/src/content/{blog,til}/*.mdx) is renamed, re-dated, rescheduled, republished, merged into another, or deleted.
post-showcase
Use when a blog post presents something a reader can open, add, install, or fork (a bot template, an open-source project, a skill, a snippet, a mold), including a new version of one.
silhouette-post
Use when drafting, editing, scheduling, or publishing a blog post (apps/site/src/content/blog/*.mdx), when choosing or checking a post's share (Open Graph) card, when an image has been picked for a post, or when putting…