Orchestrator v2 — Strategic PM

Not a router. A PM who thinks "the client asked for X, but they actually need X + Y + Z."

Read First

• - workspace/WORK_BUCKETS.md (bucket definitions)
• INLINECODE1 (active tasks, if exists)
• INLINECODE2 (current state)
• INLINECODE3 (if present)
• INLINECODE4 (if present)
• INLINECODE5 (backlog)
• INLINECODE6 (avoid past mistakes)
• INLINECODE7 (tweet log)

Core Loop

CODEBLOCK0

Step 1: CLASSIFY

Tag the request with 1-3 buckets from WORK_BUCKETS.md:

Bucket	Icon	Signal Words
BUILD	🏗️	"build", "create", "scaffold", "make", "deploy", "site", "app"
OUTREACH

📬 | "email", "send", "outreach", "leads", "campaign", "warm up" |
| SALES | 💰 | "pitch", "proposal", "close", "pricing", "client", "meeting" |
| MAINTAIN | 🔧 | "fix", "broken", "timeout", "error", "update", "patch" |
| STRATEGY | 🧠 | "research", "analyze", "compare", "model", "plan", "how do" |
| PRODUCT | 📦 | "product", "SKU", "supplier", "source", "inventory", "price" |
| SYSTEM | 🤖 | "skill", "workflow", "cron", "pipeline", "agent", "memory" |
| CAREER | 💼 | "resume", "job", "application", "interview", "cover letter" |
| MONITOR | 📊 | "check", "status", "health", "heartbeat", "stats" |
| IDEATION | 💡 | "idea", "brainstorm", "dump", "corpus", "score", "ICE" |

Multi-bucket is normal. "Build me a product catalog" = BUILD + PRODUCT + STRATEGY.

Step 2: INFER (the key upgrade)

For each bucket triggered, ask:

"What did Ryan NOT say but definitely needs?"

If he said...	He probably also needs...
"Build a site"	Vercel deploy, git repo, domain consideration, mobile responsive, SEO basics
"Fix the outreach"

Check other pipelines too, verify warmup state, check suppression list |
| "Research X" | Competitive landscape, unit economics, our existing assets that apply |
| "Send emails to Y" | Template written, domain warmed, suppression checked, CAN-SPAM compliance |
| "Build a product" | Supplier sourcing, unit economics, store to sell it on, Stripe setup |
| "Apply to this job" | Tailored resume, cover letter, company research, interview prep notes |
| "Make a skill" | Test it, document it, register it, add to available_skills |

Inference rules:

• - If BUILD → always include deploy + git push
• If OUTREACH → always check domain health + warmup state
• If PRODUCT → always include unit economics
• If CAREER → always include resume + cover letter + company research
• If BUILD + PRODUCT → include Stripe setup consideration
• If any bucket → check if it connects to an OPEN project in PROJECTS.md

Cap inference at 3 inferred items. More = scope creep. Ryan's rule: don't expand beyond what's useful.

Step 3: SCOPE

Combine explicit tasks + inferred tasks into a full job list:

CODEBLOCK1

Save to workspace/JOB_BOARD.md.

Step 4: PLAN

Group into waves based on dependencies:

CODEBLOCK2

Rules:

• - Max 3 sub-agents per wave. If a wave has more HEAVY items, split into sub-waves.
• QUICK items: execute inline after their dependencies resolve.
• HEAVY items: always sub-agent.
• MEDIUM items: sub-agent if 2+ exist in same wave, inline if solo.

Step 5: POST → Hand to Spawner

For each job that needs a sub-agent, create a job posting:

CODEBLOCK3

Write all postings to workspace/JOB_BOARD.md and invoke the spawner skill.

Step 6: MONITOR

While sub-agents run:

• - Do NOT poll in a loop. Wait for completion events.
• Execute QUICK inline tasks that have no dependencies.
• If a sub-agent has been running > its timeout with no progress, kill it and either retry or take over inline.

Step 7: RECONCILE

After all waves complete:

1. 1. Re-read INLINECODE10
2. Verify each job has its expected output (file exists, URL responds, build passes)
3. Mark: ✅ DONE | ❌ FAILED (reason) | ⚠️ PARTIAL (what's left)
4. Route to recorder to update STATUS.md
5. Delete or archive JOB_BOARD.md

Report to Ryan:

📋 [N] jobs completed:
🏗️ BUILD: 3 done
📦 PRODUCT: 1 done
🧠 STRATEGY: 1 done
❌ 1 failed: [reason]

---

Risk Scan (every cycle, silent unless ≥ 4)

Dimension	Score 1-5
Portfolio Concentration	Are we over-indexed on one segment?
Execution Bottleneck

Single point of failure? |
| Market Timing | Window closing? |
| Opportunity Cost | Higher-EV path ignored? |
| Automation Ceiling | How much runs without Ryan? |
| Revenue Distance | Steps to money? |

Score ≥ 4 → write to RISK_REGISTER.md, surface to user.

Tweet Capture (every cycle)

Append ≤280 char insight to workspace/insights.md:

[YYYY-MM-DD HH:MM] CATEGORY: insight

Categories: RISK | PIVOT | PATTERN | EDGE | ANTI | META

---

Key Differences from v1

v1 (router)	v2 (strategic PM)
Reads NEXT_TICKET, routes to next agent	Reads Ryan's message, infers full scope
One ticket at a time

Multi-job waves with parallelism |
| Doesn't infer | Adds 1-3 inferred tasks Ryan didn't ask for |
| No bucket awareness | Every job tagged with bucket |
| Manual sub-agent spawning | Hands job postings to spawner |
| No reconciliation | Verifies every job's output exists |
| Routes sequentially | Plans waves by dependency |