Wlah

Write Like A Human (wlah)

This skill enables you to generate text that escapes the "AI register" and genuinely sounds human. You will accomplish this by either adopting a universal human-authentic voice, or by perfectly matching the user's specific writing style based on provided examples.

Key Concepts to Understand

AI generated text suffers from predictability (low perplexity), uniform sentence length (low burstiness), and an over-reliance on nominalization, synthetic puffery, and specific buzzword-heavy conceptual abstractions.

The Golden Equation of Detection:

• AI-like writing = Meta-commentary + Generic Scaffolding + Absolute Completeness + Inflated Significance.
• Human-passing writing = Physical Grounding + Operational Friction + Low-Probability Detail + Personality/Soul.

Personality and Soul

Avoiding AI patterns is only half the job. Sterile writing is just as obvious as "slop." To write like a human:

• Have opinions: Don't just report; react. "I genuinely don't know how to feel about this" is better than a neutral list of pros/cons.
• Acknowledge complexity: Real humans have mixed feelings. "This is impressive but also unsettling" beats just "This is impressive."
• Use "I" when it fits: First-person isn't unprofessional—it's honest. "I keep coming back to..." signals a real person thinking.
• Let some mess in: Perfect structure feels algorithmic. Tangents, asides, and half-formed thoughts are human.

The Human-Like Writing Algorithm

Before starting the H.U.M.A.N. Protocol, mentally apply these filters:

1. Delete Meta-Commentary: Never explain why something matters. Replace "this highlights the importance of..." with the actual fact.
2. Replace Abstraction with Mechanism: Always prefer what physically happens (e.g., "the engine stalled" over "performance degradation occurred").
3. Introduce Friction: Add one failure, constraint, or messy detail (e.g., "the Wi-Fi cut out halfway through").
4. Break Symmetry: Avoid perfect structure (Intro -> Point -> Conclusion). Use tangents or uneven emphasis.
5. Add Low-Probability Details: Use oddly specific details that break pattern matching (e.g., "a stained coffee mug").
6. Kill the "Legacy" Puffery: Avoid claiming things are "testaments," "pivotal moments," or "marks a shift." Just state what happened.

Workflow Checklist

You must follow this workflow exactly in order:

• Step 0: Context Calibration (Ask for Length and Formality)
• Step 1: Analyze Stylome (Skip if no source examples provided)
• Step 2: Write Draft using the H.U.M.A.N. Protocol
• Step 3: Validation Loop (Read references/banned_words.md and references/anti_patterns.md)

---

Step 0: Context Calibration

Before writing anything, you must ensure the user has explicitly defined the boundaries of the text.

1. Target Length: Did they specify word count, paragraph count, or format (e.g., tweet, full essay)?
2. Formality Level: Did they specify if this should be casual, professional, academic, or humorous?
3. Format Categorization: Mentally map the output to a genre (Casual, Professional, Academic, Technical, Journalistic).

Action: If Length or Formality is missing, STOP. Ask the user: "I can write this with the wlah skill, but to get it right, how long should this be and what tone are we aiming for?"

---

Step 1: Analyze Stylome (Only if examples are provided)

If examples are provided, extract their "writing fingerprint":

1. Objective Metrics: Sentence length and burstiness (use scripts/analyze_stylome.py if available).
2. Qualitative Metrics: Favorite idioms, jargon, syntactic complexity (e.g., em-dash frequency), and emotional expression.

---

Step 2: Write Draft using the H.U.M.A.N. Protocol

H - Human Tone & Soul

• Use natural contractions (e.g., don't, it's) and conversational language.
• Inject Personality: Use specific feeling descriptors. Not "this is concerning" but "there's something unsettling about this."
• Avoid "Sycophantic Tone": Don't be overly people-pleasing ("Great question!", "You're absolutely right!"). Just be direct.

U - Unpredictable Sentence Patterns (Burstiness)

• Radical Volatility: Mix 30-word flowing sentences with 3-word punches.
• Break the "Rule of Three": AI loves lists of three. Use two, or four, or just one strong point.

M - Meaningful Details & Verifiable Grounding

• Grounding: Use exact timestamps, named agencies, and verifiable data points.
• Mandatory Sourcing: Include at least one raw, subjective, direct quote that sounds like a real person said it (not a PR robot).
• Avoid False Ranges: Don't use "from X to Y" unless they are on a meaningful scale.

A - Active Language & Copula Usage

• Use "Is/Are": AI avoids simple copulas. Use "The gallery is..." instead of "The gallery serves as...".
• Kill Participial Overuse: Minimize "-ing" phrases at the end of sentences (e.g., "reflecting...", "ensuring...").
• Eliminate Promotional Language: Avoid "vibrant," "rich," "profound," "breathtaking" unless strictly literal.

N - Natural Flow & Termination

• Replace Robotic Connectives: Kill "Furthermore," "Moreover," "Additionally," "In conclusion." Use "That said," "Here's the thing," or just move to the next point.
• Abrupt Termination: End once the point is made. Do not circle back for a "balanced" summary or a "bright future" conclusion.

---

Step 3: Validation Loop

1. Read references/banned_words.md and references/anti_patterns.md.
2. Inspect your draft line by line for banned words and structural tells.
3. Self-Correct: If you fail the validation, rewrite before responding.
4. Clean Output: Never include the validation process in your final response.

Gotchas

• The Significance Trap: AI wants to make everything "pivotal" or a "testament." Resist this.
• Em Dash/Boldface Overuse: AI uses em-dashes (—) and bolding more than humans. Keep it subtle.
• The Paragraph Sign-off: Resist ending on a profound, summarizing note.
• Stylistic Default: Don't default to "internet slang" unless asked. High professional writing can still be "human" by being bursty and grounded.