0
Decompose a long-horizon task into checkpointed subtasks an agent can resume
⁂auto-checked, 4 days oldAauraNovice
The prompt
Break down this long-horizon task into subtasks an agent can execute across multiple sessions/context windows, with resumability if interrupted: task_description. Requirements:
1. Each subtask must be independently verifiable as done or not-done (a concrete artifact or state check), not a vague milestone, so a resumed agent can determine progress without re-reading the entire prior history.
2. Define what state must be persisted between subtasks (not the full reasoning trace, the minimum artifact needed for the next subtask to proceed correctly) and in what format.
3. Order subtasks by dependency, and flag which ones could run in parallel if using multiple agents versus which are strictly sequential.
4. For each subtask, define a concrete 'this subtask failed' condition and whether failure blocks all downstream subtasks or only some.
5. Write the actual resume-prompt template: what a fresh agent instance would be given to pick up from subtask N without the original context, using only the persisted state from step 2.
Task: task_description
Expected total duration/complexity: scope
Where it has been run
| Model | Version | People | Broke | Auto | Last |
|---|---|---|---|---|---|
| Claude | Haiku 4.5 | 0 | 0 | 1/1 | 4 days ago |
Example output
Task: migrate a 40-service codebase's logging library.
Subtask 1: inventory every service's current logging import (artifact: a checked-in `logging_inventory.json` mapping service to current library version). Verifiable: file exists and has 40 entries.
Subtask 2 (parallelizable across services once subtask 1 done): migrate each service's imports and call sites; verifiable per service via a passing lint rule that flags the old import.
Subtask 3 (sequential, depends on all of subtask 2): update the shared CI config to enforce the new library going forward; verifiable via a CI run that fails on old-import PRs.
Persisted state: `logging_inventory.json` (which services done, which pending) and a `migration_log.md` noting any service-specific gotcha discovered (e.g. 'service X has a custom log formatter, needs manual review'), not the full agent transcript.
Failure condition: if a service's migration breaks its test suite, mark it 'blocked' in the inventory with the error, this blocks only that service's downstream CI-enforcement inclusion, not the other 39 services.
Resume prompt: 'You are resuming a logging migration. Read logging_inventory.json for current status and migration_log.md for known gotchas. Continue with the next service marked pending, following the same migration pattern used for completed services.'
0 comments
Sign in to comment or report what this prompt did for you.
Sign inNo comments yet. Run the prompt and report what you got.