-
Notifications
You must be signed in to change notification settings - Fork 0
Closed
Description
Problem
During the v0.3.0 QA session, API errors (500 Internal Server Error, 529 Overloaded) caused ~15 minutes of delays. When errors occur mid-triage:
- Progress is lost (which emails were already processed)
- User has to re-orient ("what's going on?")
- Sub-agent results may be partially completed but not captured
Proposed Solution
Add checkpoint/resume capability to the morning triage workflow:
1. Progress Tracking File
Write triage progress to a scratchpad file after each step:
# ~/.config/gws/morning-session.yaml (or scratchpad)
session_start: "2026-01-28T11:30:00"
step: "guided_triage"
processed_ids:
- {id: "19c03ccbb5b0a4b1", action: "labeled_and_archived", label: "14 - PRIVACY"}
- {id: "19c03d469a0abd5d", action: "archived"}
auto_handled:
- {id: "19c02c2f11a6db5b", action: "noise_archived"}
pending_review:
- {id: "19c0354c4e4b272d", category: "REVIEW", priority: 2}
triage_results_cache: "/path/to/triage-results.json"2. Resume on Error
If the session is interrupted:
- On next
/morningrun, detect the checkpoint file - Ask: "Resume previous session (N items remaining) or start fresh?"
- If resume: skip already-processed emails, continue from checkpoint
3. Triage Agent Results Cache
Triage agents write their output to a scratchpad file in addition to returning it. If the main agent crashes after triage but before presenting results, the cached output can be re-read.
Impact
- Eliminates progress loss on API errors
- Better UX during outages
- Enables longer triage sessions without risk
Complexity
High — requires changes to SKILL.md workflow, scratchpad file management, and resume logic.
Related
- Morning skill v0.3.0 QA session findings
- API errors observed: 500 (Internal Server Error), 529 (Overloaded)
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels