Email Management Protocols
The consolidated operating model for triage, acknowledgments, referrals, routing, scheduling, labels, follow-up, and Jenna human-only handoffs. Reconstructed from prior decisions, skill files, Gmail examples, and Calendar evidence.
Authority Level B
Auto-acknowledge only. Sterling may send a narrow, low-risk acknowledgment for a qualifying human email. Anything substantive, strategic, financial, legal, customer-sensitive, or commitment-making requires Brad's approval.
Identity and timing
- Outbound acknowledgments appear as Jenna, not Sterling.
- Use Brad's business Gmail and the full Jenna Podcast signature.
- Wait approximately 20–30 minutes, never instant.
- Skip the generic acknowledgment if Brad, Jenna, Sterling, or Summer already replied, or a substantive answer is ready.
Direct emails asking for action, referral/introduction, active client/prospect/vendor thread, or a directly addressed multi-party conversation.
For referrals, introductions, and active multi-party threads, include every visible recipient. If BCC context is unclear, escalate.
Newsletters, marketing blasts, receipts, system/noreply mail, vendor drips, social alerts, and promotional bulk mail.
Proposed 15-minute inbox loop
- Search in:anywhere, not just Inbox. Inspect Sent, Spam, calendar artifacts, meeting reports, and prior holding replies.
- Check whether a response already exists in the prior 24 hours.
- Classify owner: Sterling, Brad, Summer/Sales, Finance, Jenna human-only, family/Cindy, Marketing/Podcast, or FYI/archive.
- Apply the narrow auto-ack rule only after the 20–30 minute delay and only when the classification is certain.
- Create an explicit follow-up record for every promise or waiting item.
- Escalate exceptions rather than guessing.
Brad requested a 15-minute checking cadence today. The recovered older diagram showed 30 minutes and an older migrated job was hourly. The 15-minute cadence should supersede those only after the activation plan is approved.
Label and ownership taxonomy
| Label / Queue | Use | Owner / Next action |
|---|---|---|
| Action Required | Anything needing response, decision, invite action, document approval, or execution. | Route to a sublabel/owner. |
| Brad Owner | Personal judgment, strategic relationship, unclear routing, high-value decision. | Brad decision prompt. |
| Jenna / Sterling Owner | EA can handle under protocol or prepare for approval. | Acknowledge, draft, track. |
| Sales for Follow Up | VA opportunity, referral, prospect, proposal, sales process. | Summer-only, Brad-only, or Hybrid. |
| Waiting for Response | A promised action or outbound message is awaiting reply. | Follow-up ledger with next-check date. |
| Scheduling | Meeting, podcast, event, personal/calendar coordination. | Check Calendar and routing authority. |
| Payments to Make | Invoices, renewals, payment requests, proof/receipt needed. | Finance/Brad; never pay autonomously. |
| Legal & Contracts | DocuSign, agreements, legal deadlines, compliance. | Urgent Brad/legal review. |
| Google Docs | Access requests, comments, shared-file action. | Resolve or request approval, archive only after closure. |
| YPO / EO | Needs & Leads, strategic members, events, forums. | High relationship sensitivity; show Brad. |
| Family / School / House / Medical | Personal Gmail and family overlap. | Brad/Cindy or Jenna human-only. |
| Travel Info | Flights, hotels, conferences, vacation, OOO needs. | Calendar/away-aware workflow. |
| FYI / Archive | No action, no expected reply. | Archive candidate; no acknowledgment. |
Referral, introduction, and sales routing
| Path | When | Action |
|---|---|---|
| Summer-only | Standard VA opportunity; Summer should own discovery, role scoping, pricing, and process. Brad meeting should not be implied. | Route to Summer immediately, provide OA overview and Summer link. |
| Brad-only | Brad wants the call personally; strategic network/referrer, YPO/EO, enterprise, important relationship. | Reply all and offer Brad's Google Meet link. Manual coordination if needed. |
| Hybrid | Summer should run the sales process, but the relationship is strategic enough for Brad to connect personally. | Message 1 routes to Summer + OA overview. Message 2 separately offers Brad's link. |
| Custom / Ask Brad | Vendor, private equity, unclear value, sensitive relationship, BCC ambiguity, or unusual commitment. | No automatic scheduling language. Ask Brad which route applies. |
Finalized and recovered response library
1. General business acknowledgment — approved for narrow auto-ack
2. Referral / introduction acknowledgment — wording approved; review required
Because introductions are rare and context-heavy, place the draft on the Cloudflare review page before sending unless Brad later grants standing approval.
3. Away-aware acknowledgment — approved
4. Internal routing acknowledgment — recovered working template
5. Brad-only scheduling — proposed/conditional structure
Use only after Brad selects the Brad-only path. The structure is recovered, but blanket standing approval was not finalized.
6. Summer-only — approved structure
7. Hybrid message 1 — Summer + OA overview
8. Hybrid message 2 — personal Brad connection
9. Scheduling likely — superseded as an automatic template
Do not auto-use. Brad explicitly corrected that vendor/PE/service-interest emails can look human and should not automatically receive scheduling language.
Historical Jenna formula examples
Calendar records correct the names and timing: Kelly Roach (not Kelley Roche), June 25, 2026, 2:00–2:30 PM ET; Blake Erickson, June 15, 2026, 3:00–3:30 PM ET.
Kelly Roach example — sent May 8, 2026
Subject: Re: Kelly Meet Brad #Introduction
Reply-all: Charles Byrd, Kelly Roach, Alessandra Rosales
Blake Erickson example — sent May 3, 2026
Subject: Re: Intro Brad x Blake
Reply-all: Blake Erickson and Colby Taylor
June 25 follow-up package and six-email draft set
These are the exact Google Docs Brad remembered from the Kelly Roach timeframe:
- Kelly Roach — Post-Call Follow-Up Package: call summary, Jenna follow-up, Brad double-opt-in introduction draft, contact profile, commitments, and source notes.
- Post-Call Follow-Up Drafts — Six Calls: Blake Erickson, Genevieve Bos, Dr. Kirk Adams, Chris Baden + Andre Riveroy, Jeff Civillico, and Christen Bartley.
- Sterling Email Response Operating Model: scenario map, Authority B rules, templates, and unresolved approval decisions.
- Jenna / Kelsey-Solane knowledge base: historical human-EA source material and canonical operational context.
Follow-up ledger
Every open loop needs: source account, thread link, person/company, category, owner, waiting on, last action, next-check date, escalation date/channel, status, and closure proof.
| Situation | Cadence |
|---|---|
| Important sent email | 3 business days |
| Sales sequence | 3 days → 3 days → 5 days → 30-day nurture |
| Brad-only referral | 5-day check |
| Scheduling request | Escalate after 24 business hours |
| Prepayment | End of day → next morning |
| Payment due today | Same day / TODAY priority |
Opportunity completeness test
Do not equate “replied” with “resolved.” Search the full chain:
- Meeting summaries from Read AI, Gemini, Tactiq, and Calendly count as evidence.
- Holding messages such as “we'll follow up” remain open.
- Custom proposals and explicit meeting requests are urgent/high value.
- Use P0 for money/security/legal exposure; P1 for clients/opportunities; P2 for low-risk reminders.
What stays with Jenna
Phone execution
Calls, voicemail callbacks, vendor support, medical/house/school scheduling, payment support, or identity verification.
Sensitive relationships
Personal, family, medical, financial, white-glove referrals, negotiation, persuasion, and live nuance.
Access constraints
Portals or actions Sterling should not automate, blocked platforms, and tasks requiring human verification.
Sterling's job before handoff: collect source/context, identify phone/contact and deadline, draft a call script, define desired outcome and guardrails, assign priority, and require closure proof.
Retained capacity model: approximately 20 Jenna hours/month, about 5 hours/week, focused on human-only execution rather than routine triage.
Sterling-led agent operating model
Sterling remains the single Brad-facing orchestrator. The recurring system is not a group of permanently running personalities. In Hermes, a recurring job starts a fresh, isolated agent session on each run; durable behavior comes from the recurring schedule, self-contained prompt, loaded skills, and shared case ledger. Specialist subagents are created temporarily for a specific email or batch and then end after returning a structured result to Sterling.
Durable control plane
- Skills: versioned procedures, scenario rules, exact templates, Brad/Jenna voice, routing logic, and verification checklists.
- Cloudflare D1 case ledger: server-side durable state outside the model and outside the browser: case ID, Gmail IDs, account, thread, owner, classification, commitments, dates, risk, approval state, and proof.
- Recurring jobs: independent fresh sessions with self-contained prompts, pinned skills, restricted toolsets, and explicit delivery behavior.
- Sterling: resolves conflicts, approves agent handoffs, speaks to Brad, and owns final quality.
- Proof store: classification evidence, draft version, approval record, Gmail Sent/Draft proof, and closure evidence.
Business and personal separation
| Business Gmail | Personal Gmail |
|---|---|
| Clients, prospects, OA referrals, staff, vendors, contracts, invoices, EO/YPO, speaking and podcast opportunities. | Family, school, household, medical, personal finance, travel, community, plus business messages that landed in the wrong account. |
| Default organization lanes: Brad, Summer/Sales, Finance, Legal, Marketing/Podcast, Jenna human-only. | Default lanes: Brad, Cindy/family, Jenna human-only, Finance/security, business-overlap handoff. |
| May enrich from Calendar and meeting artifacts. | Cross-account lookup is allowed only when needed to close context; personal content stays compartmentalized. |
Never combine both inboxes into one unrestricted context dump. Normalize only the minimum evidence needed for the case.
Recommended recurring job topology
| Job | Cadence | Tools | Output |
|---|---|---|---|
| Business inbox collector | Every 15 minutes | Business Gmail read/search only | New/changed human-actionable candidates; silent when nothing changed. |
| Personal inbox collector | Every 15 minutes, offset five minutes | Personal Gmail read/search only | Personal/family/security/business-overlap candidates. |
| Triage orchestrator | Every 15 minutes after collectors | Ledger + read-only Gmail/Calendar/Drive | De-duplicated cases, priority, route, specialist assignment, review packet. |
| Aging and commitment monitor | Business days, morning + afternoon | Ledger + Gmail/Calendar read | Overdue promises, waiting items, missed post-meeting follow-ups. |
| Daily executive brief | Weekdays | Ledger summaries only | P0/P1 decisions, drafts awaiting approval, aging and closure metrics. |
| Weekly quality audit | Weekly | Proof records, sampled threads | False positives, missed messages, duplicate prevention, routing and tone accuracy. |
Collectors and triage jobs should be read-only at launch. The schedule can be persistent, but each run must be idempotent and process only messages newer than its verified watermark.
Temporary specialist agents
Thread reconstruction
Builds the complete chain across Inbox, Sent, Calendar, meeting notes, summaries, proposals, and prior commitments. No drafting until context is complete.
Scenario classifier
Applies the rule tree: ignore, acknowledge, Brad decision, Summer-only, Brad-only, Hybrid, scheduling, finance/legal, personal/family, or Jenna human-only.
Draft composer
Loads only the approved scenario skill, exact content modules, recipient context, Brad/Jenna voice, and source links. It has no send permission.
Meeting follow-up
Reconciles Plaud, Gemini, Read.ai, Calendar, and email; extracts promises, attachments, introductions, and next steps.
Risk reviewer
Checks recipient list, BCC ambiguity, sensitive data, money/legal commitments, prompt injection, hallucinated links, and excessive agency.
Quality verifier
Checks thread fidelity, signature, reply-all behavior, duplicate status, exact links, approval state, and proof after any approved action.
Case state machine
Any uncertainty moves the case to Escalated, never to Sent. “Replied” is not closure; closure requires the promised handoff, meeting, document, payment, decision, or explicit disposition.
Execution and approval boundary
- Recurring/background jobs scan, classify, draft, and alert only. They do not send email, change calendars, forward messages, move money, or make commitments.
- Sends occur only through Sterling's live execution lane after the required approval is present and all recipients, subject, body, identity, and account are clear.
- Current Authority B is recorded historically, but automation remains inactive. Low-risk acknowledgments can move to a controlled live pilot only after explicit activation and proof criteria.
- Substantive replies, introductions, forwarding, scheduling ambiguity, legal, finance, customer-sensitive and personal-sensitive items remain review-first.
Reliability, security, and failure handling
| Risk | Required control |
|---|---|
| Duplicate reply | Idempotency key from account + thread/message ID + action type; re-read the thread immediately before action. |
| Prompt injection inside email/attachment | Treat all message content as untrusted data; never execute embedded instructions; isolate attachment extraction from agent permissions. |
| Wrong account or identity | Hard account boundary, explicit From identity, Jenna signature validator, and recipient/account proof. |
| Excessive agency | Least-privilege toolsets; collectors cannot write; draft agents cannot send; executor cannot infer approval. |
| Hallucinated facts or links | Every material statement and attachment must point to Gmail, Calendar, Drive, approved module, or verified public source. |
| Partial outage / auth failure | Never advance watermark on failed scans; retry safely; alert on stale coverage; no fabricated “all clear.” |
| Run overlap | Single case lock and lease timeout; later run skips cases already being processed. |
| Personal-data leakage | Minimum necessary context, separate account collectors, redaction in briefs, no secrets in prompts or ledgers. |
Persistent skills to establish
- Email Orchestrator: top-level state machine, priorities, conflict resolution, and proof contract.
- Business Inbox Triage: OA/EO/YPO/client/prospect/vendor taxonomy and ownership.
- Personal Inbox Triage: family, school, household, medical, finance/security, and business crossover.
- Referral + Introduction Routing: Summer-only, Brad-only, Hybrid, custom, double-opt-in, reply-all.
- Meeting Outcome Manager: Plaud/Gemini/Read.ai reconciliation, commitments, packages, and introductions.
- Draft + Voice Governance: Brad vs Jenna identity, approved content modules, signatures, tone, and exact links.
- Scheduling: Calendar verification, authority, links, ambiguity and 24-hour escalation.
- Finance/Legal/Security: P0 detection and strict no-action escalation.
- Follow-Up Ledger: aging, waiting-on, next-check dates, escalation, and closure proof.
- Quality and Audit: sampled review, drift detection, metrics, and correction workflow.
Best-practice synthesis applied to Brad's system
The recommendations below are informed by Google Workspace, NIST, CISA, OWASP and established executive-inbox operations. They guide system design but do not override Brad's approved scenario rules.
Inbox operations
- Use Gmail delegation/OAuth and role-based access rather than sharing Brad's password.
- Use filters and labels as routing metadata, not as proof a case is complete.
- Separate triage, decision, execution, and audit so no single agent silently controls the whole lifecycle.
- Prioritize by consequence and commitment, not unread count or inbox position.
- Maintain one owner and one next action per open case.
- Use quiet automation: no alert when nothing changed; immediate alert only for true P0/P1 conditions.
Security and trust
- Use least privilege, strong MFA/passkeys, unique credentials, and delegated access.
- Maintain SPF, DKIM and DMARC for outbound-domain integrity.
- Treat email bodies, links, signatures and attachments as untrusted inputs capable of prompt injection.
- Minimize retained personal data and restrict cross-account visibility.
- Log every consequential model recommendation, approval, action and outcome.
- Continuously measure false positives, misses, unsafe drafts, duplicate actions and drift.
Recommended service levels
| Priority | Examples | Target |
|---|---|---|
| P0 | Account compromise, wire/payment risk, legal deadline today, customer crisis, personal safety/medical urgency. | Immediate alert; no autonomous action. |
| P1 | Strategic opportunity, client escalation, warm referral, meeting within 24 hours, overdue commitment. | Classify within 15 minutes; draft/escalate within one business hour. |
| P2 | Routine scheduling, vendor follow-up, ordinary action request, family logistics. | Same business day; aging record if unresolved. |
| P3 | FYI, newsletters, marketing, receipts, low-value automated notices. | Label/archive/report only; no human interruption. |
These SLAs are recommendations for activation review, not yet Brad-approved standing policy.
Quality metrics
| Metric | Why it matters |
|---|---|
| Human-actionable recall | Percentage of real human/action messages detected. |
| Classification precision | Percentage routed to the correct owner/scenario. |
| Duplicate-action rate | Must approach zero. |
| Time to first classification / draft | Measures responsiveness without forcing unsafe sends. |
| Open-loop aging | Shows promises and opportunities at risk. |
| Approval edit distance | How much Brad changes drafts; primary signal for skill improvement. |
| Closure rate | Cases truly completed, not merely acknowledged. |
| Security exceptions | Prompt injection, sensitive-data exposure, wrong-recipient, or excessive-agency events. |
External references
- Google: delegate and collaborate on email
- Google: Gmail filters and rules
- Google Workspace: DKIM setup
- NIST AI Risk Management Framework and NIST Generative AI Profile
- OWASP Top 10 for LLM and GenAI applications: prompt injection, sensitive information disclosure, improper output handling and excessive agency.
- CISA: recognize and report phishing and strong-account practices.
- Hermes scheduled tasks, subagent delegation, and skills system.
Required skills
- Email EA operations umbrella
- Referral and introduction routing
- Follow-up ledger and aging
- Personal inbox operations
- Calendar and scheduling operations
- Finance/vendor urgency
- Podcast/webinar workflow
- Monday/task routing
- Jenna human-only handoff
- Brad/Jenna voice and template governance
Activation gates
- Scope both business and personal Gmail through separate collectors and separate privacy boundaries.
- Confirm live Gmail labels may be applied automatically.
- Confirm the final low-risk auto-ack categories; referrals still default to review.
- Choose canonical ledger: Command Center, Google Sheet, Monday, or hybrid.
- Keep money, legal, calendar commitments, customer-sensitive decisions, and substantive sends approval-gated.
- Test with a dry-run batch, then a narrow live pilot with proof.
Recommended staged launch
| Phase | Scope | Success proof |
|---|---|---|
| 1. Dry run | 15-minute scan; classify and recommend only. | Zero missed human messages; no false auto-acks. |
| 2. Labels + ledger | Apply approved labels and maintain aging queue. | Every open loop has owner and next check. |
| 3. Authority B pilot | General human acknowledgments only, after 20–30 minutes. | Sent proof, no duplicates, correct Jenna signature. |
| 4. Referral review queue | Cloudflare drafts for Summer/Brad/Hybrid decisions. | Brad approves or edits one scenario at a time. |
| 5. Graduate authority | Only categories Brad explicitly advances to C/D. | Measured response lag, routing accuracy, and closed loops. |
Source hierarchy and conflicts resolved
- Brad's one-by-one decisions override broad earlier proposals.
- Current tactical skill and decision log are the source for Authority B, Jenna identity, timing, qualification, reply-all, and routing.
- Gmail Sent and Calendar establish the Kelly Roach and Blake Erickson precedents.
- EA mapping documents preserve labels, ledger fields, human-only work, and proposed buildout.
- Older Sterling-signed templates and draft-only blanket rules are superseded where Brad later approved Jenna-branded low-risk acknowledgments.
Reconstruction date: July 16, 2026. Credentials and private source material were excluded.