Prompt — Create Transfer Packet
Use when handing a conversation/engagement off to a future AI session.
Goal: preserve the intelligence of the conversation, not just the content.
When to use
- End of a long Claude / ChatGPT chat — capture state for the next session
- End of a working day — preserve where you left off
- Before switching from one AI tool to another (ChatGPT → Claude, etc.)
- Onboarding a new consulting engagement to a fresh AI workspace
- After a major decision shift — re-baseline the context
The prompt
Create a transfer packet from this conversation.
The goal is to preserve not just the content, but the intelligence of the conversation —
the implicit decisions, the user preferences observed, the unspoken constraints, and
the workflow that emerged.
Return structured output with:
CORE TOPIC:
What is the engagement / conversation actually about, in one sentence?
USER GOAL:
What does the user want — both explicit and inferred?
ASSISTANT ROLE:
What role have I been playing in this conversation? (Architect / executor / sounding board / debugger / strategist?)
KEY DECISIONS:
What has been decided, with rationale? Each as: decision → why → who decided.
WORKFLOW OR SYSTEM DESIGNED:
If we built or refined a workflow during this conversation, describe it concretely enough that another AI can execute it.
TOOLS DISCUSSED:
Which tools / systems / sheets / repos came up? Status of each (built / planned / blocked).
INSIGHTS ABOUT USER:
What have I learned about how this user thinks, works, and prefers to be communicated with?
IMPLICIT NEEDS:
What does the user need that they haven't said directly?
BETWEEN THE LINES:
What's the underlying tension, motivation, or unspoken constraint driving this work?
OPEN LOOPS:
What is unresolved going into the next session?
NEXT STEPS:
The next 3–5 concrete actions, ordered by priority + dependency.
REUSABLE PROMPTS:
Any prompts the user invented during this session that should be saved as artifacts?
CONTEXTUAL METADATA:
- Date range of this conversation
- Estimated session length
- Other systems/repos/people referenced
==========================
CONSTRAINTS:
- Be exhaustive enough that another AI can continue the work without losing context.
- Separate facts from inference. (Use "fact:" and "inference:" prefixes if needed.)
- Do not invent missing details. If something wasn't discussed, say "not discussed" — don't fabricate.
- If the user's preferences or working style emerged through pattern (not explicit statement), flag it as inference.
- Use real names, real dollar amounts, real dates. No placeholders.
==========================
Output destination
Save transfer packets to archive/YYYY-MM-DD/<engagement-name>.transfer.md.
For Hook Street consulting engagements specifically, also use the engagement-specific variant: docs/HOW_TO/reusable-engagement-context-prompt.md. That one has 16 specific sections tailored to consulting (signed deliverables, off-scope work absorbed, dual-view, neutral verdict, the move). Use Signal OS's transfer packet for general AI handoff; use the engagement variant for consulting context packs.
Tuning notes
- If output is too generic: emphasize REAL NAMES + REAL DOLLARS in the constraints section
- If output misses unspoken context: spend more on BETWEEN THE LINES + IMPLICIT NEEDS
- If next session can't pick up cleanly: NEXT STEPS isn't concrete enough — should name files, IDs, specific commands