Resume After Context Loss

This is an operational strategy that agents receive via .thrum/strategies/. It describes the exact sequence to follow when resuming after a context loss event such as compaction, a new session start, or a crash.

If you are resuming after context loss (compaction, new session, crash, etc.), follow this sequence exactly. Do not skip steps.

Step 1: Re-Register with Thrum (Mandatory)

thrum quickstart --name <agent-name> --role <your-role> --module <branch-name> --intent "Resuming <task-or-epic>"
thrum inbox --unread
thrum sent --unread
# Tip: thrum inbox --unread peeks without marking read; thrum message read --all to mark all read

You must re-register even if you were registered in the previous session. Registration is per-session and does not persist across context loss.

Step 2: Orient from Beads

bd show <EPIC_ID>              # What is the overall state of the epic?
bd list --status=in_progress   # Is anything currently mid-flight?
bd ready                       # What tasks are unblocked and available?

Step 3: Orient from Git

git --no-pager log --oneline -10   # What was committed recently?
git status                          # Is there any uncommitted work?
git diff                            # What changes exist but are not staged?

Step 4: Pick Up from the First Incomplete Task

Read beads status and git history to determine exactly where work stopped. Then:

• If a task is in_progress in beads and has uncommitted changes: review the diff, decide whether to complete or reset, then continue
• If a task is in_progress but git shows it is fully committed: close the task in beads and move to the next
• If all visible tasks are completed: verify nothing was missed, then proceed to the next unblocked task or wrap up

Key principle: DO NOT redo completed work. Trust beads task status and git commit history. If a task is closed in beads and committed in git, it is done — do not re-implement it.

Next Steps

• Thrum Registration — the registration protocol that Step 1 above depends on
• Context Management — save context before session end so resumption is faster after a loss event
• Beads and Thrum — how task tracking and messaging work together to make context recovery possible
• Agent Coordination — the full session template including start, during, and end-of-session protocols