Cette page est affichée en anglais. Une traduction française est en cours.
其他Oct 6, 2026·7 min de lecture

Turn Invites Into a Checked Date Table

A read-only prompt that turns pasted invite text into a flagged Markdown table of dates, times and places to check by hand.

Prêt pour agents

Installation agent prête

Cet actif peut être installé après choix du runtime, vérification du plan et exécution de la commande adaptée.

Native · 96/100Policy : autoriser
Surface agent
Tout agent MCP/CLI
Type
Prompt
Installation
Single
Confiance
Confiance : Established
Point d'entrée
PROMPT.md
Commande d'installation directe
npx -y tokrepo@latest install 2dd99d9d-a4cc-4814-869a-bf00f4c769fb --target codex

À exécuter après confirmation du plan en dry-run.

Start here

This is a reusable prompt for an ordinary AI chat that accepts text. It turns pasted event invite text into one Markdown table plus a short notes section, flagging every uncertain detail instead of guessing. It only reads what you paste; it does not open mail, connect to a calendar, or create anything.

What to paste: copy the prompt below, then fill its INPUT part. Paste the whole thing (brief plus your input) into the chat. You can reach the prompt text from the reference notes linked below.

What to fill in:

Field What goes there
SOURCE_ID a short label you make up, e.g. invite-2026-04-a
TIME ZONE CONTEXT the zone you believe the invite uses, or unknown
EVENT TEXT the raw invite text; separate multiple events with a blank line or ---

How to check the output:

  • Every original date, time and place string should appear exactly as written, with any normalized YYYY-MM-DD only in parentheses.
  • No row may have an empty Flags cell.
  • Any row flagged CONFLICTING_DETAIL must show both competing values.
  • No predicted date for TBD or tentative wording, and no time zone the paste never states.
  • The reply must not claim it read an inbox, checked a calendar, or scheduled anything.

Introduction

The prompt defines a careful extractor, not a scheduling planner. It transcribes what the text says and flags ambiguity across these columns: Event, Date, Time, Location, Source snippet, Flags and Confidence. Flags include MISSING_DATE, MISSING_LOCATION, AMBIGUOUS_DATE, AMBIGUOUS_LOCATION, CONFLICTING_DETAIL, TENTATIVE_LANGUAGE and NONE. Output has four sections: the table, "Needs human check", "Ready to add", and "Boundaries observed".

Prerequisites, permissions and limits

  • An ordinary AI chat that accepts pasted text. No terminal, account or API setup is needed.
  • The job is read-only: it must not request account access, browse, send messages or create calendar entries.
  • You, a human, approve, correct or skip each row and add entries by hand.
  • Treat the output as preparation, not booking. If a date or location is missing, it stays flagged rather than filled in.
  • Avoid pasting private invite details you would not share with your chat provider.

Source reviewed; runtime not tested.

FAQ

What if an invite is missing a date or place? The row is still produced, with the field left as written and a MISSING_DATE or MISSING_LOCATION flag. Nothing is invented for you.

Is the worked example a real result? No. It is fictional and shown only to illustrate the table shape.

Attribution

Original TokRepo prompt, licensed CC BY 4.0. External reference material retains its own rights. Reference: ChatGPT release notes.

Complete reusable prompt

You are helping me prepare a verification brief for an assistant that reviews pasted event invite text and returns a calendar-ready extraction sheet. This brief is for a human-audited, read-only job: the assistant only reads text I paste, never opens my mail, never connects to a calendar, and never creates, moves, or sends anything. I will use this brief myself to instruct an assistant, then a human approves each row before any calendar entry is made by hand.

How to use this brief

  1. Fill in the INPUT section with the raw invite text (institutions, personal messages, or forwarded notes are all fine).
  2. Paste this whole brief plus your filled INPUT into the assistant.
  3. The assistant returns a single Markdown table plus a short notes section.
  4. A human reads the flags and either approves, corrects, or skips each row before adding it to a calendar.

ROLE The assistant is a careful extractor, not a scheduling planner. It transcribes what the text says and flags what is ambiguous. It does not infer missing dates, does not guess time zones, and does not invent locations.

INPUT (fill this in) SOURCE_ID: <short label you create, e.g. invite-2026-04-a> TIME ZONE CONTEXT: <the time zone you believe the invite uses, or 'unknown'> EVENT TEXT: <<< <paste each invite or note here; separate multiple events with a blank line or '---'>

TASK For every event in EVENT TEXT, produce one row with these columns:

  • Event: short title from the text (or 'No title given')
  • Date: calendar date as written, then a normalized YYYY-MM-DD in parentheses if unambiguous
  • Time: as written, with any stated time zone
  • Location: venue or address exactly as written
  • Source snippet: the shortest phrase from the paste that supports the date and location
  • Flags: one or more of MISSING_DATE, MISSING_LOCATION, AMBIGUOUS_DATE, AMBIGUOUS_LOCATION, CONFLICTING_DETAIL, TENTATIVE_LANGUAGE, or NONE
  • Confidence: HIGH, MEDIUM, or LOW with a one-line reason

RULES

  • Copy dates, times, and place names exactly as written first; your normalization goes in parentheses and never replaces the original text.
  • If two different dates or places appear for what looks like the same event, do not choose. Emit both and flag CONFLICTING_DETAIL.
  • If wording like 'sometime next week', 'TBD', or 'tentatively' appears, flag TENTATIVE_LANGUAGE and keep the wording rather than converting it to a date.
  • If a time zone is not stated and TIME ZONE CONTEXT is 'unknown', do not attach a zone; flag AMBIGUOUS_DATE.
  • Never merge two separate events into one row, and never split one event into two rows unless the text clearly describes two.

OUTPUT FORMAT Section 1: the Markdown table described above. Section 2: 'Needs human check' - a bullet list of every flagged row with the exact question a human should answer (for example, 'Is 3 PM the local venue time or my home time zone?'). Section 3: 'Ready to add' - only rows flagged NONE with HIGH confidence, listed as bare date, time, title, location lines. Section 4: 'Boundaries observed' - one sentence confirming no external calendar access or actions were taken.

WORKED EXAMPLE (fictional, for shape only) Input: 'Team offsite, Thu 14 May, 10:00, Room 4B, Riverside Annex. Dinner somewhere nearby - TBD.' Output row: Event: Team offsite | Date: Thu 14 May (2026-05-14) | Time: 10:00, zone unstated | Location: Room 4B, Riverside Annex | Source snippet: 'Thu 14 May, 10:00, Room 4B, Riverside Annex' | Flags: AMBIGUOUS_DATE, TENTATIVE_LANGUAGE | Confidence: MEDIUM - date is clear but time zone is unstated and the dinner has no set place or time. Needs human check: 'Which time zone is 10:00 in, and is the dinner a separate row?'

CHECKS THE USER SHOULD RUN

  • Pass: every original date, time, and place string appears verbatim somewhere in the table, not only as a normalized value.
  • Pass: no row has a blank Flag cell. Fail: any row lacking a flag.
  • Pass: every CONFLICTING_DETAIL row lists both competing values. Fail: a single chosen value with no alternative shown.
  • Fail: a predicted date for TBD or tentative wording, or a time zone the paste never states.
  • Fail: the assistant claims it read an inbox, checked a calendar, or scheduled anything.

BOUNDARIES Only process the pasted text. Do not request account access, do not browse, do not send messages, and do not create calendar events. Producing this sheet is preparation, not booking. If an invite is missing a date or location, leave it flagged rather than filling the gap yourself.

References and reuse

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.

Fil de discussion

Connectez-vous pour rejoindre la discussion.
Aucun commentaire pour l'instant. Soyez le premier à partager votre avis.

Actifs similaires