Resume From Ledger

Resume From Ledger

Recover from interrupted long-running tasks by reading the persistent TODO ledger. Pairs with the session-resilience rule.

When to Trigger

User says any of:

• "continue [the] audit/refactor/migration"
• "resume [from where we left off]"
• "pick up the X task"
• "what's left on X"
• References a task that hit usage limits

Procedure

Step 1: Locate ledger

ls -lt ~/.omc/plans/*-todo.md 2>/dev/null | head -10

If multiple, ask user which task. If one obvious match, confirm before proceeding.

Step 2: Parse ledger

• Count [x] (done), [ ] (pending), [!] (failed)
• Identify next pending item

Step 3: Status report (BEFORE doing work)

Resuming: <task name>
Done: M / N items
Failed: K items (will retry: yes/no)
Next: item-X (<description>)

Wait for user "go" / "skip failures" / "retry K first".

Step 4: Execute next batch

Follow the rules in rules/session-resilience.md:

• Max 3-5 items per batch
• Mark [x] immediately after each item
• Mark [!] for failures with reason
• Stop after batch, report progress, await go-ahead for next batch

Failure Recovery

If item-X failed previously:

1. Read its previous attempt notes (in ledger or /tmp/)
2. Diagnose root cause
3. Propose new approach
4. Confirm with user before retrying

Anti-Patterns

• ❌ Re-running already-done [x] items (wastes tokens)
• ❌ Re-planning from scratch (defeats the ledger's purpose)
• ❌ Skipping the status report (user loses visibility)
• ❌ Auto-retrying failed items without diagnosis