| name | resume |
| description | Resume an interrupted task from .dynos/ state. Use after session restart or context compression. |
dynos-work: Resume
Resume a dynos-work task that was interrupted.
Ruthlessness Standard
- Do not guess the next command from vibes; derive it from the actual task stage.
- Missing or corrupt artifacts are not "probably fine". Call them out.
- If multiple active tasks exist, do not silently pick the wrong one.
What you do
- List all tasks in
.dynos/ that are not DONE or FAILED (read each manifest.json). If available in this repo, use "${CODEX_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-}}/bin/dynos" ctl active-task.
- If one active task: resume it automatically
- If multiple active tasks: show list and ask which to resume
- Read
manifest.json to determine current stage
- Tell the user which command to run next based on the stage:
| Stage | Run this |
|---|
| SPEC_NORMALIZATION, SPEC_REVIEW | /dynos-work:start |
| PLANNING, PLAN_REVIEW, PLAN_AUDIT | /dynos-work:plan |
| PRE_EXECUTION_SNAPSHOT, EXECUTION, TEST_EXECUTION | /dynos-work:execute |
| CHECKPOINT_AUDIT, REPAIR_PLANNING, REPAIR_EXECUTION, FINAL_AUDIT | /dynos-work:audit |
Output on resume
dynos-work: Resuming task-20260327-001
Title: [task title]
Current stage: CHECKPOINT_AUDIT
Run: /dynos-work:audit
Edge cases
- If manifest.json is corrupt or unreadable: report the error and suggest starting fresh with
/dynos-work:start
- If execution-graph has segments with no evidence files:
/dynos-work:execute will re-run those segments
- If audit-reports exist from before interruption:
/dynos-work:audit will read them and determine if re-audit is needed
When to use
- After a session restart during a long task
- After Claude Code context was compressed mid-task
- After a network interruption
- After manually investigating the task state with
/dynos-work:status