- Update all skill SSH commands from sshpass to key-based auth (~/.ssh/archipelago-deploy) - Add proxy_connect_timeout 120s to nginx Ollama location blocks - Add new polish/sweep skills for overnight automation - Add demo content (documents, photos) for demo stack - Add .ssh/ to .gitignore Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
3.8 KiB
3.8 KiB
Skill: Production Polish (Overnight Orchestrator)
Main entry point for the Archipelago production polish plan. Reads plan.md at the project root, determines the current week based on today's date, and executes the tasks for that week.
How It Works
- Read
plan.mdfrom the project root - Determine the current week from the schedule:
- Week 1: March 10–16 — Silent Failures & Error Handling
- Week 2: March 17–23 — Loading States & Visual Feedback
- Week 3: March 24–30 — Form Validation & Input Quality
- Week 4: March 31 – April 6 — Backend Robustness
- Week 5: April 7–13 — WebSocket & Real-Time Quality
- Week 6: April 14–20 — Deployment & Infrastructure Hardening
- Week 7: April 21–27 — Accessibility, Polish & Edge Cases
- Week 8: April 28 – May 4 — Integration Testing, Final Sweep & ISO
- Execute tasks for the current week, in order
- After completing tasks, run
/sweepto verify - Deploy and verify with
/deploythen/check-server
Execution Flow
Step 1: Read the plan
Read plan.md and find the current week's section
Step 2: Check what's already done
Run the verification checks for the current week's tasks. For example in Week 1:
- Count remaining
.catch(() => {})patterns - Count remaining
console.logoutside dev guards - Count remaining
unwrap()in backend production code - Check if hardcoded credentials still exist
Step 3: Work on the next incomplete task
Pick the first task in the current week that still has violations (hasn't met its acceptance criteria). Fix violations one file at a time:
- Read the file
- Apply the fix following the pattern described in the task
- Verify the fix compiles/type-checks
- Move to the next violation
Step 4: Verify after each batch of fixes
After fixing all violations for a task:
- Frontend:
cd neode-ui && npx vue-tsc --noEmit - Backend:
ssh archipelago@192.168.1.228 "cd ~/archy && cargo check" - Run the task's specific acceptance grep/check
Step 5: Deploy when a task is complete
When all violations for a task are fixed and verified:
./scripts/deploy-to-target.sh --live
Then verify:
ssh -i ~/.ssh/archipelago-deploy archipelago@192.168.1.228 "systemctl is-active archipelago && curl -s http://localhost:5678/health"
Step 6: Move to the next task
Repeat Steps 3-5 for the next incomplete task in the current week.
Step 7: When all tasks are done
Run /sweep for a full quality report. If clean, the week is complete.
Rules
- Never change functionality — only improve quality of existing code
- Never change the design — use existing glassmorphism classes, color tokens, and layout patterns
- Always deploy after changes — don't leave undeployed code
- Always verify after deploy — check server health
- Build Rust on the dev server — never compile Rust on macOS
- Commit after each completed task — atomic commits with
fix:orrefactor:prefix - If something breaks, revert — don't push forward with broken code
Arguments
If $ARGUMENTS is provided:
week N— Force execution of week N regardless of datetask N.M— Execute only task N.M (e.g.,task 1.3)status— Show completion status for all weeks without executingsweep— Run sweep only, no fixes
Example Usage
/polish # Auto-detect week, work on next incomplete task
/polish week 1 # Force Week 1 tasks
/polish task 1.3 # Work on just task 1.3
/polish status # Show what's done and what's left
/polish sweep # Just run the quality sweep
For Overnight TUI
Launch with:
/loop 30m /polish
Each 30-minute cycle:
- Checks current week
- Finds next incomplete task
- Fixes as many violations as possible in the time available
- Deploys and verifies
- Reports progress