Rehydrating a New Session

~25 min · rehydration, new-session, sources

Level 0Window Watcher

0 XP0/50 lessons0/13 achievements

0/100 XP to next level100 XP to go0% complete

Rehydration is selective loading

A new session should not blindly load everything. It should read the durable boot files, the live handoff board, the exact files in scope, and the latest relevant history entry if needed. The goal is to reconstruct the working state, not the entire emotional weather of the previous session.

Verify current truth, then act

After reading handoff, inspect the repo. Current code beats old notes. Current status beats old plans. Current human instruction beats stale docs. The first action of a new session is to align reading-knowledge with running-knowledge.

The cwkPippa rehydration ritual

This codebase has a real ritual: read NEW-SESSION-REMINDER.md, read sibling CLAUDE.md, hold TODO.md mapping, read vault entry. Five minutes of disciplined loading saves hours of cold-context flailing. The ritual is documented in memory feedback_session_start_routine; it exists because skipping it has cost real time.

Code

Session rehydration order·yaml

rehydration_order:
  1_bootstrap:        "identity and global rules"
  2_project_overlay:  "repo CLAUDE.md and instructions"
  3_live_board:       "current operating state"
  4_target_files:     "current code/content in scope"
  5_recent_history:   "only if needed"
  6_verify:           "git status, tests, routes"

First-action checklist·text

Before any edit:
  - run `git status` to see real local state
  - run any quick test that proves the spine still passes
  - confirm the human's first message matches the live board's 'next'
  - if mismatch, ask before drifting

External links

OpenAI conversation state guide

Exercise

Write a rehydration checklist for a future AI instance taking over your project. Include which files to read first and which old materials are only references.

Hint

If your checklist is more than ten items, it is a research project, not a rehydration.