Chatgpt Custom Instructions

# Export format: ChatGPT custom instructions

ChatGPT's custom-instructions feature exposes two fields, each with a
~1500-character soft limit:

1. **Field 1** — "What would you like ChatGPT to know about you to provide
   better responses?"
2. **Field 2** — "How would you like ChatGPT to respond?"

## Output language

**Render all user-visible labels and section titles in the user's
conversation language** — not the language this template file is written
in. The English labels below are canonical references for maintainers;
translate them at render time. The two ChatGPT field names themselves
("Field 1: About me" / "Field 2: How to respond") should also be
translated when shown to the user. Body content inherits its language
from the source files.

For example, a Chinese-speaking user sees `字段 1：关于我` instead of
`Field 1: About me` in the export display, and the section structure
inside each field uses Chinese (e.g., `请：` / `永不：` rather than
`Do:` / `Never:`).

## Mapping rules

**Field 1 — About me** should contain context (facts), drawn from:
- Role pack: identity, responsibilities & goals, tools & stack
- Project pack: summary, current status, key decisions (condensed),
  rejected options (condensed — the "don't re-suggest X" value lives here)

Write in third-person-about-self or first-person-statement voice:
> "I am an AI product manager at a fintech startup. I'm building…"

**Field 2 — How to respond** should contain instructions, drawn from:
- Role pack: collaboration preferences, taboos

Write in imperative mood, directed at ChatGPT:
> "Respond concisely with tradeoffs. Never suggest X. Prefer bullet lists…"

## Length handling

- Target 1200–1400 chars per field (leave headroom; users often edit).
- If output exceeds 1500 chars, do NOT auto-truncate. Instead:
  1. Show the user both fields with exact character counts.
  2. Flag which field is over and by how much.
  3. Suggest concrete cuts, using these per-field compression orders
     (items listed first should be compressed first; items listed last
     should be preserved longest):

     **Field 1 — About me:**
     stack details → rejected options (name-only, drop the "why not")
     → decisions (name-only, drop the "why") → current status
     → responsibilities → identity

     **Field 2 — How to respond:**
     stack details → collaboration preferences (merge similar bullets)
     → identity restating → taboos

     Never drop a taboo silently; if forced to compress them, tell the
     user which got cut.

  4. Let the user pick what to cut. Do not decide for them.

## Output format

Present the export as two clearly-separated code blocks so the user can copy
each into the respective ChatGPT field: