| name | resume-context |
| description | Semantically restore current task execution state from .fusion/ recovery data. Use at the start of a new session or after context compaction. |
Fusion Resume Context (Codex)
Read the current task's .fusion/ recovery data and rebuild execution context.
When to Use
- At the start of a new Codex session (if not auto-injected by hook)
- After context compaction or memory loss
- When resuming a task after a long pause
- When another agent needs to pick up where you left off
Steps
- Run the resume script to get an overview:
python3 ./.trellis/scripts/fusion/resume.py
For full recovery data (including all JSON fields):
python3 ./.trellis/scripts/fusion/resume.py --full
- Read
.fusion/recovery.json for complete state (if exists)
- Read
.fusion/handoff.md for human-readable summary (if exists)
- Read
plan.md to cross-reference progress
- If
contract.md exists, read acceptance criteria
- Synthesize all information and output a recovery summary
- Ask user: "Where should we continue from?"
Recovery Levels
| Level | Name | Accuracy | Method |
|---|
| 1 | Exact Resume | ~100% | Native session restore (platform-specific) |
| 2 | Semantic Resume | ~85% | .fusion/ checkpoint data (this skill) |
| 3 | Cold Resume | ~50% | plan.md + task.json + git history only |
Output Format
The resume script outputs:
- Task title and directory
- Handoff summary (if available)
- Recovery progress (current slice, next action, blockers)
- Recent git commits and uncommitted changes
Notes
- If no
.fusion/ data exists, fall back to Cold Resume (Level 3)
- The hook
fusion-session-start.py auto-injects recovery data at session start
- This skill provides manual/on-demand recovery when auto-injection is insufficient
