Skip to content
OpenAgentSkill
Browsing is open · Catalog captured 2026-11-21 · Automatic updates and account services are paused.
Development

tweet-draft-reviewer

Review tweet drafts in Claude Code against 8 voice rules. Scores 1-10, breaks down every rule, and rewrites anything that scores below 7.

BrianRWagner · Static checks on file

Installation

Read the source and dependencies before running a command. Copying does not install or execute anything.

npx skills add BrianRWagner/ai-marketing-claude-code-skills --skill tweet-draft-reviewer

About this skill


name: tweet-draft-reviewer description: Review tweet drafts in Claude Code against 8 voice rules. Scores 1-10, breaks down every rule, and rewrites anything that scores below 7.

Tweet Draft Reviewer

Paste a tweet draft and get a score out of 10, a rule-by-rule breakdown, and a rewrite if the score is below 7. Takes 30 seconds. Saves you from posting something that sounds like a chatbot wrote it.

Built on 8 voice rules distilled from real content analysis — what separates high-engagement tweets from the ones that get skimmed.


Mode

Detect from context or ask: "Just a score, full review, or full review with alternate versions?"

ModeWhat you getBest for
quickScore out of 10, the single weakest rule, and a one-line fixLast check before posting
standardScore, all 8 rules broken down, and a rewrite if it scores below 7Any draft you care about
deepFull review plus 3 rewrites with different hooks, and a note on which rule you break most often across the batchBatch reviews, or tuning your voice over time

Default: standard. Use quick if they say "just score it." Use deep for folder scans or when they ask for options.


How to Use

Single draft:

Review this tweet draft: [paste tweet here]

Batch scan:

Review all tweet drafts in my content/tweet-drafts/ folder.

Skill Instructions (for Claude Code)

When this skill is invoked, follow these phases exactly.


PHASE 1: INTAKE

Determine the input type:

Input A — Direct paste: User provided draft text inline. Proceed to PHASE 2 with that text.

Input B — Folder scan: User asked to review drafts folder. Run:

VAULT="${VAULT_PATH:-$(pwd)}"
find "$VAULT/content/tweet-drafts" -name "*.md" 2>/dev/null | while read f; do
  if ! grep -q 'reviewed: true' "$f" 2>/dev/null; then
    echo "UNREVIEWED:$f"
  fi
done

If no vault path was given and no content/tweet-drafts/ exists in the current directory, ask:

Where is your tweet drafts folder? (full path, e.g. $HOME/vault/content/tweet-drafts)

Mode C — Ambiguous: No draft provided and no folder context. Ask:

Paste your tweet draft here, or tell me the path to your tweet-drafts folder and I'll scan it.

PHASE 2: ANALYZE

Apply all 8 rules to the draft. For each rule, record ✅ PASS or ❌ FAIL with a one-line reason.


The 8 Rules

Rule 1: No "I" opener

  • FAIL: First word is exactly I (standalone — not "In", "It", "If")
  • PASS: Anything else

Rule 2: Strong opener

  • FAIL: First sentence ends with ? OR starts with "Have you", "Do you", "Are you", "What if", "What would"
  • PASS: Declarative statement, specific number/fact, named scenario, or emotional setup

Rule 3: No AI tells

  • FAIL: Contains any of: delve, certainly, game-changing, game changer, it's worth noting, invaluable, unleash, revolutionize, transformative
  • PASS: None of those words detected

Rule 4: No generic closers

  • FAIL: Ends with (or contains near the end): what do you think, drop a comment, thoughts?, let me know in the comments, agree?, sound familiar?
  • PASS: Ends with a statement, directive, punchline, or thread hook

Rule 5: Corey Test (specificity)

  • FAIL: Uses vague language without specifics — "it changed how I work", "massive results", "so much better" — no numbers, names, or concrete outcomes
  • PASS: Contains at least one specific: a number, timeframe, named tool, or concrete result

Rule 6: Character count

  • PASS: 280 characters or fewer
  • THREAD PASS: Over 280 chars BUT sections are numbered (1/, 2/, 3/) or clearly separated with line breaks — count as PASS
  • FAIL: Over 280 chars with no thread formatting

Rule 7: Single point

  • FAIL: Makes 3+ distinct unrelated claims with no clear through-line
  • PASS: One core idea, even if supported by 2–3 details

Rule 8: Punchy rhythm

  • FAIL: Any sentence over 20 words OR preamble like "I've been thinking a lot about..." / "Something I've noticed recently is..."
  • PASS: Short sentences, no preamble, gets to the point by line 2 at latest

PHASE 3: OUTPUT

Print this exact format:

TWEET REVIEW
────────────

Score: X/10

Rule-by-Rule:
1. ✅/❌ No "I" opener — [reason]
2. ✅/❌ Strong opener — [reason]
3. ✅/❌ No AI tells — [reason]
4. ✅/❌ No generic closers — [reason]
5. ✅/❌ Corey Test — [reason]
6. ✅/❌ Character count — [reason + actual count]
7. ✅/❌ Single point — [reason]
8. ✅/❌ Punchy rhythm — [reason]

────────────

Scoring table (passes → score):

PassesScore
810/10
79/10
68/10
56/10
45/10
34/10
23/10
11/10
00/10

If score ≥ 7:

→ Ready to post. Use Typefully or your scheduler to queue it.

If score < 7 — add a rewrite:

Suggested Rewrite:

[rewritten tweet that fixes all failing rules]

PHASE 4: SELF-CRITIQUE

Before delivering the output, check:

  1. All 8 rules evaluated — Count the rule lines in your output. There must be exactly 8. If any are missing, add them.
  2. Score matches pass count — Count the ✅ marks. Verify the score matches the table above. Correct it if not.
  3. Rewrite quality — If a rewrite was provided, check it against all 8 rules mentally. Does it pass the ones the original failed? If the rewrite would score below 7 itself, rewrite it again.
  4. Reason quality — Each rule result must have a specific reason, not a generic one. "Too vague" is not enough — say what specifically is vague.

Fix anything that fails before delivering.


Example Review

Input:

I've been thinking a lot about how AI is really changing the way founders 
approach content creation and productivity, and I think it's worth noting 
that the teams who figure out async AI workflows are going to have a 
massive advantage. What do you think about this?

Output:

Read the complete source ↗