# AI Utilization, Scope, and Handoff Histogram Date: 2026-05-17 Source: Live a02 files under `/opt/veritize-docker/ops` Window: Last 7 days of inbox, outbox, approval, and MCP files ## Executive Summary The AI mesh is operating as a file-and-relay workflow, not a single unified agent. SC is the highest-volume execution/evidence lane. SGPT, SG, and SGrok are review/advisory lanes with different scopes. CC/CGPT are coordination and deployment lanes. VIRA is a gate/report lane. The biggest operational imbalance is that SC carries the bulk of evidence production, while CC receives many assignments and CGPT performs many direct production closures. SGrok is active, but its current relay prompt is scoped to press/strategy and can reject infrastructure/status work unless the dispatch is phrased inside that role. ## Agent Scope | Agent | Primary Scope | Current Operating Pattern | Notes | |---|---|---|---| | SC | Execution, probes, health checks, evidence reports, queue automation | Highest-volume producer; writes platform health, test summaries, APAR/PTF evidence, status sweeps | Watcher/session alerts exist, but live tmux/process checks showed SC sessions and watcher running | | SG | Architecture, security, adversarial review, HIPAA/HL/TPM risk review | Reviews plans and dispatches; flags risk aggressively | Correctly treats private HL/Rosalind/WK boundaries as hard constraints | | SGPT | Engineering/risk review, full-stack implementation review, AGE status reports | Responds reliably through relay; active status confirmed | Best fit for code/design review and implementation risk | | SGrok | Press, strategy, UX, value filtering, public messaging | Active but role-prompt is narrow; rejects generic ops/status templates | Use for UX/market narrative/value-filter work, not infra verification | | CC | Control/coordination queue, dispatch intake, sequencing, governance bridge | High inbox count, lower direct output count | Used as operator/coordination lane, not primary evidence producer | | VIRA | Approval/contact queue/gate reports | Scheduled daily contact queue reports | Gate lane, not a task executor | | CGPT | Codex execution/deployment/documentation lane | Direct production work and written completion records | Current interactive operator lane | ## Workload Histogram File-count based utilization from the last 7 days. ### Assigned Work Scale: max = 39 assigned inbox files ```text CC 39 | ######################################## SC 36 | ##################################### SGPT 23 | ######################## SGrok 22 | ####################### SG 14 | ############## VIRA 0 | CGPT 0 | ``` ### Outputs / Evidence Scale: max = 130 output/evidence files ```text SC 130 | ######################################## CGPT 41 | ############# SGPT 35 | ########### SGrok 32 | ########## SG 26 | ######## CC 9 | ### VIRA 7 | ## ``` ## Utilization Table | Agent | Assigned Inbox | Outputs/Evidence | Outbox | Approval | MCP | Output Bytes | |---|---:|---:|---:|---:|---:|---:| | SC | 36 | 130 | 1 | 118 | 11 | 585,887 | | SG | 14 | 26 | 15 | 6 | 5 | 440,244 | | SGPT | 23 | 35 | 25 | 6 | 4 | 378,093 | | SGrok | 22 | 32 | 22 | 4 | 6 | 151,400 | | CC | 39 | 9 | 0 | 7 | 2 | 13,588 | | VIRA | 0 | 7 | 0 | 7 | 0 | 889 | | CGPT | 0 | 41 | 0 | 37 | 4 | 94,356 | ## Agent-To-Agent Handoff Histogram Counts are mention/reference based across recent files. They show who is being used by whom, not exact API token spend. ```text SC -> CC 83 | ######################################## SC -> SGPT 40 | ################### SGPT -> SC 31 | ############### SGrok -> SC 28 | ############# CC -> SC 27 | ############# SC -> VIRA 27 | ############# SGrok -> CC 26 | ############ CC -> SGPT 23 | ########### SG -> SC 22 | ########### SGPT -> CC 21 | ########## SG -> CC 20 | ########## CC -> VIRA 18 | ######### SGPT -> VIRA 17 | ######## SC -> SG 13 | ###### SGPT -> SG 12 | ###### SG -> SGPT 11 | ##### SGrok -> SGPT 11 | ##### CC -> SG 11 | ##### SG -> VIRA 10 | ##### SGrok -> VIRA 8 | #### CGPT -> CC 8 | #### CC -> SGrok 7 | ### SGrok -> SG 6 | ### CGPT -> SC 6 | ### CGPT -> SGPT 5 | ## SC -> SGrok 3 | # SG -> SGrok 2 | # SGPT -> SGrok 2 | # SGPT -> CGPT 2 | # CGPT -> SG 2 | # CGPT -> VIRA 2 | # ``` ## Current Tasking Snapshot | Agent | Current / Latest Tasking Evidence | Status Read | |---|---|---| | SC | `/opt/veritize-docker/ops/mcp/SC_NEXT_JOB.md` and latest SC reports | Assigned to AGE prod-status, platform health, DataCert/APAR evidence work | | SG | `/opt/veritize-docker/ops/mcp/SG_DISPATCH_NOW.md`, `SG_OUTBOX_AUTO.md` | Active, but flags private-data scope risks strongly | | SGPT | `/opt/veritize-docker/ops/mcp/sgpt/SGPT_INBOX.md`, `SGPT_OUTBOX.md` | Active and responsive | | SGrok | `/opt/veritize-docker/ops/mcp/SGROK_DISPATCH.md`, `SGROK_OUTBOX.md` | Responsive, but role-scoped to press/strategy | | CC | `/opt/veritize-docker/ops/mcp/CC_DIGEST.md`, queue inbox | Coordination and backlog sequencing | | VIRA | Daily `VIRA_REPORT_CONTACT_QUEUE_*` | Gate/report only | | CGPT | `CGPT_DONE_*` and prod wakeup records | Interactive execution/deployment lane | ## Recommendations 1. Keep SC as the evidence/probe runner, but watch for overload. Its output count is 3x the next lane. 2. Use SG only for architecture/security/risk review, especially HL/TPM/HIPAA boundaries. 3. Use SGPT for code review, implementation plans, and risk checks before deployment. 4. Use SGrok for UX, messaging, value filtering, and public narrative. Do not send it generic infra-status templates unless its system prompt is updated. 5. Use CC as the sequencing/control lane. Do not expect CC to produce the bulk of technical evidence. 6. Keep VIRA as the formal gate/report lane. 7. Continue keeping WK, HL, and Rosalind segmented. The current AI mesh repeatedly references those boundaries, which is correct. ## Evidence Files Created By This Inventory - Local script: `ai_utilization_inventory_2026_05_17.py` - Remote JSON: `/tmp/ai_utilization_inventory_2026_05_17.json` - This report: `AI_UTILIZATION_SCOPE_HISTOGRAM_2026-05-17.md`