agentleFS
pivoshenkopivoshenko/pivoshenko.ai/skills/blog-write/SKILL.md

blog-write

Skill6 starsChanged 44 days agoMITView on GitHub

Write and edit blog posts for pivoshenko.dev - interrogate for raw material, outline as a dependency graph, confirm sections, draft section-by-section, run the local anti-slop edit passes (references/anti-slop.md), ship as MDX. Use when the user says "write a blog post", "draft a post about X", "edit/revise/improve this article", "tighten this draft", "turn this into a post", "publish to my blog", or shares notes/an experience meant for pivoshenko.dev. The goal is bespoke practitioner writing, not generic AI prose.

We split this skill into the rules it gives an agent

8 rules. 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Never draft from a one-line idea.

    line 23
  2. Distinctive:

    No raw material for a section -> ask targeted questions, don't pad.

    line 32
  3. Distinctive:

    No thesis -> no post, only notes.

    line 37
  4. Distinctive:

    Every section needs an edge to the thesis.

    line 45
  5. Distinctive:

    Check: read points in order, circle any term/claim not yet introduced -> reorder or add setup

    line 46
  6. Very specific:

    Anti-slop: - run references/anti-slop.md: kill throat-clearing intro, summary outro, rule-of-three padding, hedge stacks, "it's not X - …

    line 63
  7. Very specific:

    Facts: - every number/command/claim traces to user, vault, or repo.

    line 65
  8. Distinctive:

    Never invent metrics, dates, quotes, features

    line 77

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Blog Write

Target: pivoshenko.dev, posts at site/content/posts/*.mdx.

Goal = bespoke content. The post must contain things only the author could write - numbers, failures, decisions, opinions. Anything the first Google result could say -> delete. Padding is where generic prose comes from.

Modes

  • Write - idea/notes -> published MDX. Full flow (1-6)
  • Edit - existing draft -> improved. Map its headings into sections (step 3), confirm, then 4-6. Present per-section changes; never silently restructure

Flow

1. Interrogate - Raw Material Before Prose

Never draft from a one-line idea. Mine the author first:

  • What happened - the incident, the build, the decision
  • Specifics - numbers, commands, configs, error messages, dates
  • The failure - what broke, what was tried and abandoned, what it cost
  • The opinion - what they believe that others don't, and why

Source order: user > vault (06 WRITING draft, project/source notes) > repo. Never invent. Vault path unknown -> ask the user for it; no vault -> the user's own message is the only source. No raw material for a section -> ask targeted questions, don't pad.

2. Thesis

One sentence the post defends. Reader finishes -> can repeat it back. No thesis -> no post, only notes. Say so and help find one.

3. Outline as Dependency Graph

Information = DAG. A section may only use concepts established by earlier sections.

  • List sections; for each write the main point (a claim, not a heading)
  • Mark dependencies: section C leans on B's claim -> B comes first
  • Every section needs an edge to the thesis; no edge -> cut
  • Check: read points in order, circle any term/claim not yet introduced -> reorder or add setup

Confirm the outline with the user before drafting. Outline is cheap, drafts are not.

4. Draft, Section by Section

Voice non-negotiables:

  • Paragraphs 1-3 lines, never more; a paragraph that grows -> split
  • Rhythm: claim, claim, claim, beat
  • First person singular; lowercase brand/tool names; posts never articles

One section at a time. Each section carries >=1 author-only artifact (number, error, config, decision + why). Section has none -> back to step 1 for that section, not into generalities.

5. Edit Passes - Separate, in Order

  1. Structure - DAG still holds after drafting drift; every section still earns its edge to the thesis
  2. Anti-slop - run references/anti-slop.md: kill throat-clearing intro, summary outro, rule-of-three padding, hedge stacks, "it's not X - it's Y"
  3. Voice - smell test: notebook entry not pitch; survives chopping in half; loses nothing if adjectives deleted
  4. Facts - every number/command/claim traces to user, vault, or repo. Untraceable -> cut or ask

6. Ship

  • File: site/content/posts/<kebab-slug>.mdx
  • Frontmatter: title, date (today, YYYY-MM-DD), description (1-2 plain sentences, no hype), tags (lowercase kebab-case, single word preferred)
  • just check from the pivoshenko.dev root must pass
  • Draft lives in the vault -> set status: published + published_url on that note; leave the rest of the note alone

Rules

  • Steps 1-3 are not skippable. A full draft delivered straight from a one-line idea is the failure mode this skill exists to prevent
  • Never invent metrics, dates, quotes, features
  • The author's opinion stays the author's. Don't sand off a strong position into "it depends"

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.