Checkpoint protocol for resuming multi-phase commands after session interruptions
Checkpoints let multi-phase commands survive session interruptions. When a phase completes, the command writes a checkpoint file. When the command is re-invoked, it reads the checkpoint and resumes from the first incomplete phase.
Before starting a multi-phase command, check context usage:
/compact to free space, or finishing the current phase, committing the checkpoint, and starting a fresh session./handoff to generate a handoff document before context runs out, so the next session can resume cleanly.Checkpoints exist precisely for this scenario — running out of context mid-work is expected, not exceptional. The command proceeds either way, but the user knows checkpoints will protect their progress.
Checkpoint files live in docs/checkpoints/:
docs/checkpoints/<command>-<ID>.md
Where:
<command> is the command name: debug, epic, feature, implement, plan<ID> is the item identifier the command is working on (e.g., FEAT-003, EPIC-001, BUG-012, the plan filename slug)Examples:
docs/checkpoints/feature-FEAT-003.md
docs/checkpoints/implement-2026-02-12-notifications.md
docs/checkpoints/debug-BUG-012.md
docs/checkpoints/epic-EPIC-001.md
docs/checkpoints/plan-FEAT-007.md
If the item has no formal ID (e.g., a debug session started from a symptom description), derive a short slug from the context (e.g., debug-login-timeout).
---