AJ — Strategic Decisions Required (Salesforce + Infra)
1. Salesforce Production Data Storage — 91.4% (6 GB headroom, weeks not months)
Choice between:
- Option A — Archive to Snowflake (2-3 days work; preserves history; one-time effort + recurring 6-monthly cycle). Ideal target: completed Booking__c records >24 months + orphan ContentVersion/Attachment.
- Option B — Buy Salesforce Data Storage add-on (~₹X/month — quote needed from AE). Fast but recurring opex.
- Recommended: A first, B as fallback if archive yields <2 GB recovered.
2. Salesforce Fullcopy Sandbox — 101.4% OVER quota
Sandbox is over → blocks Apex deploy validation.
- Option A — Refresh sandbox (production-data reseed; LOSES current dev work-in-progress; 24-48h rebuild time).
- Option B — Truncate non-essential customs (Booking__c historical, Audit__c) via
sf_bulk2_create_joboperation=delete. Preserves dev work. - Recommended: B first (preserves cycle), A only if B insufficient.
3. Salesforce Agentforce License (parked capability — 19 sf_agent_* tools)
- License procurement → External Client App →
sfap_apiscope. Cost: Salesforce AE quote. - Currently 19 tools return structured FEATURE_NOT_ENABLED (Gotcha #7 — counts as PASS, not broken).
- Decide: pursue license now (use case: agentic Booking__c workflows?) or keep parked indefinitely.
4. R-TaskFlow AWS Production Migration
Currently personal-VPS tier (5-user R&D). Migration triggers:
- Backup activation (BACKUP_ENCRYPTION_KEY + R2 bucket)
- Sentry activation (DSN provisioning)
- IaC/RDS/ALB/SES decisions
- Estimated cost: ~$200-400/month AWS (Postgres RDS + ALB + SES) vs current VPS overhead negligible.
- Decide: trigger migration, or hold at VPS tier.
5. Image Bloat Rebuild Prioritization (~12GB recoverable)
- mirofish (5.8GB), graphiti (5.65GB), hermes (4.89GB) — multi-stage rebuild pattern proven via hermes-consul (784MB).
- Effort: ~2-4h per image, sequential (avoid concurrent build pressure).
- Risk: low, reversible (keep old image tag).
- Decide priority order: graphiti first (Permanent 8 — highest reuse) seems natural; confirm or override.
Each decision above blocks ONE concrete autonomous-actionable workstream. Once decided, execution proceeds without further AJ involvement (per feedback_expert_recommendations.md).