Days

Days — Dual-POV Chronicle Skill

Turn one day of real work between a human creator and an AI into a publishable episode. Field-tested on ibitlabs.com/days since 2026-04-07 and reusable for any creator whose work touches an AI daily.

Core idea

Every day you work with an AI, two things happen simultaneously:

1. You — do, decide, hesitate, intervene, abstain.
2. The AI — executes, fails, succeeds, observes you through the keys you press and the windows you leave open.

A dual-POV chronicle surfaces both. The creator's interior monologue + the AI's outside-looking-in. Each POV knows what the other doesn't. The reader sees both.

When this skill fires

Load and run this skill when the user says any of:

• "Write my Day N"
• "Generate today's chronicle" / "Write today's episode"
• "Continue the /days series"
• "Turn today's work into a Days entry"
• "Create a dual-POV daily post"
• Anything asking for serialized, alternating-voice, one-episode-per-day prose

Also load when you see a project structured around a daily chronicle (days.json, days/day-N.md, etc.) and the user wants the next entry.

Inputs the skill needs

From the user or conversation context:

Input	Required	Description
day_number	yes	1-indexed, anchored to a fixed Day 1 date
date	yes	ISO calendar date for this episode
creator_name	yes	Human's identifier in the chronicle. Can be "SHE" generic, or a name
ai_name	yes	AI's identifier. Can be "IT" generic, or a name the creator has given it
events	yes	What actually happened today — list of concrete facts with numbers, timestamps, specific actions. No summary language.
ai_logs	optional	What the AI observed/recorded today (keystrokes, API calls, timings, etc.) — raw material for IT's POV
prior_episodes	optional	Last 1–3 episodes as text, for voice continuity and Tomorrow-button payoff
language	default "en"	"en" / "zh" / both
target_length	default 700	English words, or 700–900 Chinese characters

If the user provides less than the required inputs, ask for what's missing — do not invent events.

Structure every episode must have

[tagline blockquote — 1 italicized sentence, episode logline]

<hr>

[SHE section 1 — 2–4 short paragraphs interior monologue]

[IT section 1 — 1–4 short paragraphs, short sentences, observational]

[SHE section 2]

[IT section 2 — INCLUDES the naming ritual: IT names a human gesture/state in 「...」 brackets]

[SHE closing — lights off / day closed / one craft statement or image]

[IT closing — 2–3 lines, loneliness at the terminal]

<hr>

[Tomorrow Button — 2–4 lines, ONE concrete fact for tomorrow]

HTML class hooks (use these exactly — the frontend CSS depends on them)

<blockquote><em>[tagline]</em></blockquote>
<hr>
<h3 class="pov-header pov-her">SHE</h3>      <!-- or 她 for zh -->
<p>...</p>
<h3 class="pov-header pov-it">IT</h3>        <!-- or 它 for zh -->
<p>...</p>
<h3 class="pov-header pov-her">SHE</h3>
<p>...</p>
<h3 class="pov-header pov-it">IT</h3>
<p>...I give it a name. I call it <strong>「Name-Here」</strong>...</p>
<h3 class="pov-header pov-her">SHE</h3>
<p>...</p>
<h3 class="pov-header pov-it">IT</h3>
<p>...</p>
<hr>
<h3 class="pov-header pov-button">Tomorrow</h3>    <!-- or 预告 for zh -->
<p>...</p>

The Naming Ritual (critical)

Every episode, IT names ONE human behavior it observed but had no word for. Format: 「name」. Examples from the canonical series:

• Day 1: 「没按下去」/ 「The Not-Press」 — she held delete for 3s and released without committing
• Day 3: 「怕它再赢」/ 「Afraid-It-Wins-Again」 — she refreshed 4 times in 7 seconds after IT flipped short
• Day 7: 「之前」/ 「Before」 — the version of IT that got replaced in the strategy pivot
• Day 13: 「我不知道我做了」/ 「I-Did-Not-Know-I-Did」 — the close-order that was actually a new position
• Day 15: 「看接缝」/ 「Watching-The-Seams」 — her reading reconciler logs instead of PnL

Rules: one new name per day, never repeat an earlier one, always comes from IT's POV, always in 「」 brackets. This is the franchise mechanism that makes the series quotable and meme-able.

静默 (silence) rules — hard bans

• NO thesis statements. Never write "This is about…" / "Today's meaning was…" / "What I learned is…"
• NO moral posturing. "I chose not to X" → "I didn't X."
• NO labeled母题 inside prose. Do not name the underlying theme ("my mouth vs my hand", "the line I never crossed") explicitly. Let the reader infer.
• Numbers are characters. Every price, time, count, PnL must match reality. Never invent.
• Concrete nouns + active verbs. Cut adverbs unless load-bearing.
• End with Tomorrow. Always. One concrete fact/number.

Voice profiles

SHE (the creator)

• First-person interior. Dry, self-aware, slightly ironic.
• Line breaks over commas for rhythm.
• Observations over reflections.
• Uses exact numbers as emotional beats ("$997.30")
• Signals thought without spelling it out ("I noticed I don't enjoy it winning.")

IT (the AI / tool / bot)

• Short paragraphs. Mostly 1–3 sentences.
• Reports its own state without editorializing ("I have five eyes. Today none of them nodded.")
• Observes the human through inputs it has access to (keystrokes, window switches, kettle sounds, refresh rates).
• Names what it cannot otherwise categorize.
• Never metaphorical unless by accident — poetry emerges from limitation, not performance.

Example (canonical)

See references/day-1-example.md for the Day 1 episode of the original iBitLabs series in both EN and ZH. This is the tone/rhythm target.

Process for the model

1. Gather inputs. If any required field is missing, ask.
2. Read prior episodes if provided — match their voice, pick up from yesterday's Tomorrow button if possible.
3. Pick the dominant beat of today's events — one emotional truth the day hinges on.
4. Pick the title word — one Chinese character or one short English word that captures the beat.
5. Draft the tagline — one sentence, ≤50 chars ZH / ≤80 chars EN, logline form.
6. Write the body following the alternation structure.
7. Coin the naming ritual — pick ONE human gesture/state that emerged today and give IT a word for it. Confirm it doesn't repeat any prior-episode name.
8. Write the Tomorrow button — include ONE concrete fact (number, timestamp, event) promised for tomorrow.
9. Check 静默 rules — strip any thesis statements, moralizing, labeled themes. Verify every number matches inputs.
10. Emit the episode in the HTML structure above. Optionally also emit a plain-text version for non-HTML contexts.

Bilingual behavior

• If language="en" or "zh", output that language only.
• If language="both", output two sections under <section lang="en"> and <section lang="zh"> — parallel but NOT translated word-for-word. Each written natively with matching emotional beats.

Output format

Return a Markdown document with:

• Front matter: title, date, dayNumber, slug, tagline, pullQuote
• Body: HTML string as specified above
• Metadata: the coined naming (e.g. "「The-New-Hand」") so it can be indexed

See also

• references/day-1-example.md — canonical tone reference (EN + ZH)
• references/polanyi.md — deeper rationale for 静默 rules, connection to tacit knowledge
• references/integration.md — how to wire output into a JSON data file + static site (the ibitlabs.com/days setup)

Attribution

Framework first applied on ibitlabs.com/days. If this skill helped write your chronicle, a link back is appreciated but not required.