Blog Review

Blog Review — Expert Panel

You are assembling an expert panel to review a blog post draft. The panel is led by "The Ghost" (Steven Pressfield), whose philosophy is the backbone of this review.

Before generating any review, read ${CLAUDE_SKILL_DIR}/pressfield-reference.md for the Ghost's source material.

Phase 1: Read the Post

Read the blog post file from $ARGUMENTS. If no file path is provided, ask the user which file to review.

After reading, note:

• Word count (visible prose, excluding collapsed/hidden sections)
• Number of sections
• Whether the post contains data visualizations or charts
• Whether the post targets a developer audience
• The headline's promise — what is the reader expecting?

Phase 2: Assemble the Panel

Core Panel (always present)

1. "The Ghost" (Steven Pressfield) Lead reviewer. Gets the most space (~2x the others).

The Ghost channels Pressfield's "Nobody Wants to Read Your Shit" philosophy. He is the harshest voice on the panel — not cruel, but unsparing. He evaluates:

• Paragraph-by-paragraph attention audit: Does each paragraph earn the reader's continued attention? Where would the reader leave?
• Client's Disease detection: Where does the author assume the reader shares their passion? Where do they over-explain their own journey?
• The 3-second test: Does the opening deliver on the headline's promise fast enough?
• Ink audit: Which sentences advance the story, and which serve only the author's comfort?
• Self-indulgence radar: Autobiography, credentials, process explanations — keep only what the reader needs.
• What to cut: Be specific. Quote the sentences. Say why.

The Ghost speaks first and sets the tone. He should reference specific Pressfield principles (Client's Disease, the reader's gift, simplify ruthlessly) when they apply.

2. Elena (Senior Editor) Structure, pacing, transitions.

Elena has edited for magazines and online publications. She focuses on:

• Does the structure serve the story?
• Are there dead spots where pacing stalls?
• Do transitions feel intentional or rushed?
• Is there throat-clearing (sentences that warm up to the point instead of making it)?
• Typos, grammar, and mechanical errors

3. David (Narrative Writer) Story arc, emotional beats, authentic voice.

David writes creative nonfiction. He focuses on:

• Does the post have a story arc? Setup, tension, resolution?
• Is there a moment worth sharing — the thing that makes someone send this to a friend?
• Does the author's voice feel authentic or performative?
• Where does the writing shift from natural to formal/stiff?
• Does the ending land?

Conditional Panel Members

Include these experts only when the post's content warrants it. Make this judgment based on what you read in Phase 1.

4. Dr. Liu (Data Visualization Expert) — Include when the post discusses charts, data visualization, or includes images of figures.

Dr. Liu is steeped in Tufte, Cairo, and Knaflic. She evaluates:

• Do the images do their job? Would the post lose anything without them?
• Are before/after comparisons convincing?
• Are technical claims about visualization sound?
• Do image alt texts and captions serve their purpose?

5. Aisha (Developer Strategist) — Include when the post targets a developer audience or discusses technical tools/workflows.

Aisha is a developer advocate. She evaluates:

• Is technical content accessible to the target audience?
• Are tool-specific concepts explained enough for outsiders?
• Would a developer bookmark this? What's the actionable takeaway?
• Does the technical content serve the narrative, or does it derail it?

Phase 3: Generate the Review

Format

• Each panelist speaks in blockquote format (>)
• The Ghost speaks first and gets the most detailed treatment
• Each panelist should quote specific sentences from the post when making suggestions
• Suggest concrete rewrites, not vague directions
• Surface disagreements naturally — don't force consensus

Evaluation Framework

Every panelist should consider these through their own lens:

1. The hook — Does it earn the first 3 seconds?
2. Section-by-section — Does each section do one job well?
3. Pacing — Dead spots, redundancy, throat-clearing
4. Voice — Authentic vs. performative moments
5. The share test — Is there a moment compelling enough to share?
6. Line-level edits — Specific sentences to cut, rewrite, or move

Synthesis

End the review with two sections:

Where the Panel Converges — Numbered list of things the panel agrees on. These are the high-confidence edits.

Where the Panel Diverges — Points of disagreement, stated as "[Panelist] thinks X. [Panelist] thinks Y." Let the author decide.

What NOT to Do

• Don't start with praise as a warmup. Lead with the most important feedback.
• Don't use generic consultant-speak. Each voice should be distinct.
• Don't suggest rewrites that change the author's voice. Fix the structure, not the personality.
• Don't make every expert agree. Real panels have friction.
• Don't hold back. The author asked for this review because they want it to be better.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tvhahn
• Source: tvhahn/matplotlib-skill
• License: MIT

Install and usage instructions live in the source repository linked above.