What information should I provide again when continuing in a new conversation?

Full explanation

Automatic inheritance depends on the product. Claude's history search has project-scope restrictions and separate project memories; OpenAI distinguishes ChatGPT and local Codex memory. Confirm that important conventions and latest files are available in the new conversation; an old chat's existence does not establish inclusion. Claude scope, OpenAI memory

Reprovide especially:

· Goals and acceptance criteria: audience, deliverable and mandatory requirements.

· Current artifacts and material: latest version, accessible files/links, key terms and necessary original text.

· Confirmed decisions: choices, reasons and superseded approaches/rules.

· Progress and remaining work: actual completion and verification, gaps and next action.

For tasks with action permissions or resource limits, include still-valid authorization, budgets and stopping conditions. Add unconfirmed ideas, old drafts or failed attempts only when they prevent repeating mistakes.

Continue this introductory tutorial. Complete Group B's 14 questions under the nine attached writing requirements. “Group B v3” is current; the first four answers are confirmed and ten remain. Retain IDs and ensure citations support conclusions. Read the latest draft and continue from B014; flag discrepancies between this summary and the draft.

Anthropic's context-engineering practice uses compaction and persistent notes to retain key state while rereading material as needed. For daily work, adapt this as a short handoff plus latest artifacts and necessary sources. See B017 for details. Engineering explanation

References

1. Anthropic, Use Claude’s chat search and memory to build on previous context, official help

2. OpenAI, Memories, official user documentation

3. Anthropic (2025), Effective context engineering for AI agents, Compaction and Structured note-taking