Jake · WF4 Dashboard

YOUR ROLE

You own the strategic decisions and sign-offs. Your tasks are short but gate several downstream workstreams — prioritise the early ones.

DAY 1 PRIORITY

J1 Submit Research Preview application + J2 Approve Anthropic budget. Both are unlock-the-door moves.

IF YOU SLIP

Trung cannot start the IT build without J2 (budget) + J6 (PRD sign-off). SEO team cannot finalise the system prompt without J7 (template sign-off). Rollout gated on J8.

📥 YOU RECEIVE

Research Preview approval from Anthropic · PRD draft from Trung · Templates from Phuong Anh + An Thai · T2 validation from SEO Lead

⚙️ YOU DO

10 tasks — total 6–8h across 2–3 weeks. See task list below.

📤 YOU PRODUCE

Budget approval · Behaviour boundary decisions · Client selection · PRD sign-off · Final go-live green-light · QueryMind MCP decision

👉 CONSUMED BY

Trung (J2, J4, J6 unlock his build) · Phuong Anh (J5 picks test clients she validates) · SEO team (J7 sets template bar)

Submit Research Preview application

⏱ 1h W1

▼

Prerequisites

None — start anytime

Blocks (downstream)

Terminal — nothing waits

📋 Steps

Apply for callable_agents, memory_stores, and outcomes via Anthropic Research Preview form.
Not needed for WF4 MVP — but start Day 1 so access arrives in time for WF2.
Document application ID + submission date in ClickUp.

✅ Done when

Application submitted. Confirmation email received. Logged in ClickUp.

Approve Anthropic API budget ceiling

⏱ 15m W1

▼

Prerequisites

None — start anytime

Blocks (downstream)

📋 Steps

Approve ~$100 ceiling for Sprint 1 testing (safety margin for experimentation).
Approve ~$15/mo production budget (at 8 clients × 1 cycle × ~$0.55–0.70/run).
Confirm payment method on Claude Platform.

✅ Done when

Budget approved + payment method confirmed. Trung can start paid runs.

Audit existing external subscriptions for MCP access

⏱ 30m W1

▼

Prerequisites

None — start anytime

Blocks (downstream)

Terminal — nothing waits

📋 Steps

Verify ClickUp plan tier supports API access + MCP connector.
Verify Slack plan tier supports bot tokens + MCP connector.
Verify Google Workspace service account quota.
Verify GHL API access is enabled on the agency account.

✅ Done when

All 4 subs confirmed at sufficient tier. Any upgrades queued before Trung starts MCP setup.

Define agent behaviour boundaries for WF4

⏱ 45m W1

▼

Prerequisites

None — start anytime

Blocks (downstream)

T6T8T9T10

📋 Steps

Decide: (a) Can the agent post directly to Slack, or only draft for human approval?
Decide: (b) Can the agent update ClickUp task comments?
Decide: (c) Can the agent write files directly to client Drive folders?
Decide: (d) Can the agent auto-embed reports in GHL contact records, or manual only?
Document 4 yes/no decisions in ClickUp — each affects what Trung's orchestration script does vs what humans do.

✅ Done when

Four yes/no decisions documented. Trung can finalise orchestration script scope.

Select 2 test clients (T1 synthetic + T2 real low-risk)

⏱ 1h W2

▼

Prerequisites

None — start anytime

Blocks (downstream)

T12

📋 Steps

Pick the synthetic/sandbox fixture client for T1 (Linh will build fake data — just need a plausible shape).
Pick a real, low-risk client for T2 (established retainer, tolerant of an experiment).
Brief the real client: "Next month's report goes through a new automation — SEO Lead still reviews before you see it." Frame as improvement, not experiment.

✅ Done when

Synthetic + real client chosen. Real client pre-briefed. T2 can proceed.

Review + sign off on the PRD before build kickoff

⏱ 45m W1

▼

Prerequisites

None — start anytime

Blocks (downstream)

📋 Steps

Read Part 4 of the Action Plan doc end-to-end.
Focus on: Definition of Done (§8), scope boundaries (§1), secrets plan (§5).
Flag any scope expansions or risk tolerances you disagree with.
Sign off or send back with edits.

✅ Done when

PRD signed off. Trung starts IT build kickoff (T1 onwards).

Review Phuong Anh's template + An Thai's KPI glossary

⏱ 1h W2

▼

Prerequisites

P1A1

Blocks (downstream)

📋 Steps

Read monthly-report-template.md from Phuong Anh.
Read kpi-glossary.json from An Thai.
Judge: does this represent "a good report looks like THIS"? You have final say.
If yes, approve. If no, specific feedback to the owners before Trung mounts them on the agent.

✅ Done when

Both artifacts approved (or revised and re-approved). Agent system prompt can finalise.

Final sign-off after T2 → green-light production rollout

⏱ 30m W3

▼

Prerequisites

P6T12

Blocks (downstream)

P7L3

📋 Steps

Read Phuong Anh's T2 validation recommendation.
Review the T2 output yourself (the actual report Trung's agent produced).
Green-light or red-light the full production rollout to the remaining 7 clients.
If green: confirm rollout plan + cadence with team.

✅ Done when

Production rollout green-lit. Last gate cleared before all-client cycle.

Decide QueryMind MCP/CLI path (blocks WF2 timeline)

⏱ 1h W3

▼

Prerequisites

None — start anytime

Blocks (downstream)

Terminal — nothing waits

📋 Steps

Read Part 1 "Critical Dependency" callout — 3 options for QueryMind MCP access.
Conversation with Trung: effort estimate for Option A (build thin MCP wrapper, 3–5 days).
Decide: (a) Trung builds wrapper in WF1→WF2 gap, (b) wrap QueryMind CLI if exists, (c) defer WF2.
Commit Trung to the scope — blocks WF2 planning.

✅ Done when

Decision made + logged. Trung has scope for the WF1→WF2 gap work.

J10

Monthly cost review at end of month 1 in production

⏱ 1h Month 1 close

▼

Prerequisites

Blocks (downstream)

Terminal — nothing waits

📋 Steps

Pull Claude Console spend data for month 1.
Compare actual vs projection: ~$5/mo at 8 clients was the projection.
If variance >50%: decide whether to adjust client pricing model, throttle runs, or tune agent for fewer tokens.
Approve or flag next-workflow (WF1) go/no-go based on cost + quality confidence.

✅ Done when

Cost review signed off. WF1 go/no-go decision made.

📤 YOU HAND OFF TO

Trung (J2, J4, J6 unlock his build) · Phuong Anh (J5 picks test clients she validates) · SEO team (J7 sets template bar)

📥 YOU RECEIVE FROM

Trung (J6 PRD, J10 cost data, J9 QueryMind effort estimate) · Phuong Anh (J7 template, J8 T2 recommendation)