checkpoint

name	checkpoint
description	Save current task execution state to .fusion/ for context recovery across sessions. Use at key milestones, before context compaction, or when switching tasks.

Fusion Checkpoint (Codex)

Save the current execution state to .fusion/recovery.json and .fusion/handoff.md.

When to Use

• After completing a significant slice or milestone
• Before a long pause or task switch
• When context usage is approaching ~60%
• After making important technical decisions
• When encountering a blocker worth recording

Steps

1. Confirm there is an active task (read .trellis/.current-task)
2. Analyze current state:
   • Read plan.md to determine current slice progress
   • Read task.json for task metadata
   • Check git status for current working files
   • Collect key decisions and blockers from this session
3. Run the checkpoint script:

   python3 ./.trellis/scripts/fusion/checkpoint.py \
     --slice <current-slice> \
     --status "<current step description>" \
     --files "<files being edited>" \
     --source "<current skill name>" \
     --next "<recommended next step>"
   

4. If you need to add a blocker or decision:

   python3 ./.trellis/scripts/fusion/checkpoint.py \
     --blocker "description" \
     --decision "chose X::because Y"
   

5. Confirm save completed, show handoff.md summary to user

Output Files

File	Purpose
.fusion/recovery.json	Machine-readable execution state
.fusion/handoff.md	Human-readable session handoff summary
.fusion/events.jsonl	Append-only event log

Notes

• The Python script is platform-agnostic; it works the same in Claude Code, Codex, and other platforms
• Only the primary agent writes recovery.json; other agents are read-only
• handoff.md is fully rewritten each checkpoint (not appended)