Resume sessions

Use Nessie to resume a previous agent run in a new Claude Code, Codex, MCP, or CLI-capable agent session.

This does not reopen the old process. It gives the new agent enough source context to continue the work: the original goal, the important constraints, the latest state, and the next actions.

Where to get the ID:

1. Open Library
2. Click into the source root or session you want to resume
3. Use Copy ID in the session detail view, or two-finger click the item in Library and choose Copy ID
4. Paste that ID into the new agent session

Paste this into the new agent:

Use Nessie to resume this agent session: SESSION_OR_TRANSCRIPT_ID

Read the beginning and recent tail, write a compact takeover brief, then continue from the latest state.

The ID must be a Nessie UUID copied from Library. It can identify a transcript, node, source, or Nessie conversation.

MCP equivalent:

Call nessie_head with:
id: SESSION_OR_TRANSCRIPT_ID
lines: 5

Then call nessie_tail with:
id: SESSION_OR_TRANSCRIPT_ID
lines: 40

CLI equivalent:

If the agent has CLI access, read a small beginning window and the recent tail. The handoff state usually lives near the end, so bias toward tail:

nessie head SESSION_OR_TRANSCRIPT_ID -n 5
nessie tail SESSION_OR_TRANSCRIPT_ID -n 40

If you do not have an ID, search for the project, branch, error, file path, or topic first:

nessie grep "release blocker in billing migration" --type transcript
nessie tail TRANSCRIPT_ID -n 40

What the agent should do:

1. Read the beginning to recover the goal and constraints
2. Read the recent tail to recover current state, decisions, blockers, and next actions
3. Search distinctive terms from the tail, such as file paths, branch names, errors, people, or issue IDs
4. Write a compact takeover brief
5. Continue the requested work from that brief