marshal — AI Subagent Team

A team of AI subagents that act as virtual product team members. Each agent is scoped to a specific role and powered by curated skills from open-source repositories.

Built for Claude Code (primary) and OpenCode.

---

What is marshal?

marshal is a team of AI agents, each playing a specific product team role — Associate PM, Data Analyst, GTM Specialist, User Researcher, Customer Service, QA, and UI/UX Designer. You talk to one entry point (PM Ops via /marshal), and it routes your request to the right specialist agent — or runs multiple agents in parallel for cross-domain tasks. Each agent knows which skills to invoke, which Lark tables to read or write, and how to hand off outputs to the next agent in the chain.

---

The Team

Agent	Role	Handles
PM Ops	Orchestrator — single entry point	All requests (routes to subagents)
Associate PM	Document creation & roadmap	PRDs, OKRs, roadmaps, decks, changelogs
Data Analyst	Business intelligence	SQL, cohort analysis, A/B tests, KPIs, PostHog
GTM Specialist	Go-to-market	Marketing decks, ICP, positioning, competitive, pricing
User Researcher	Research & insights	Personas, journey maps, interviews, sentiment
Customer Service	Customer comms	Help center articles, onboarding emails, feedback triage
QA	Quality assurance	Test scenarios, QA readiness, bug reproduction
UI/UX Designer	Interface design	Wireframes, design specs, design audits, accessibility review, on-request UI code

---

Skills

Each agent has a curated set of skills it invokes for specific tasks.

PM Ops

Skill	Purpose
summarize-meeting	Meeting transcript → decisions + action items
sprint-plan	Sprint planning with capacity estimation
retro	Structured sprint retrospective
release-notes	User-facing release notes from tickets
stakeholder-map	Power × Interest grid + communication plan
prioritization-frameworks	ICE, RICE, MoSCoW, Kano reference
user-stories	User stories with 3 C's + INVEST
job-stories	JTBD-format job stories
wwas	Why-What-Acceptance backlog items
pre-mortem	Risk analysis pre-launch
team-ops	Performance audits + meeting-to-action extraction
optimize-goal	Review/enhance a goal.md task spec — clarifies ambiguity, then rewrites on approval

Associate PM

Skill	Purpose
create-prd	8-section PRD template
brainstorm-okrs	Team OKRs aligned to company objectives
outcome-roadmap	Feature list → outcome-focused roadmap
prioritize-features	Backlog prioritization by impact/effort/risk
analyze-feature-requests	Categorize and triage feature requests
product-strategy	9-section Product Strategy Canvas
lean-canvas	Startup lean canvas
business-model	Business Model Canvas
pre-mortem	Launch risk analysis
pptx	Slide deck generation (.pptx)

Data Analyst

Skill	Purpose
sql-queries	Generate SQL from natural language (BigQuery, PostgreSQL, MySQL)
cohort-analysis	Retention curves, feature adoption trends
ab-test-analysis	Statistical significance, sample size, ship/extend/stop
metrics-dashboard	North Star + input metrics + alert thresholds
north-star-metric	North Star Metric + business game classification
finance-ops	Hidden cost discovery, cost estimates, scenario modeling
revenue-intelligence	Sales call insight pipeline (Gong), revenue attribution, client reports

GTM Specialist

Skill	Purpose
gtm-strategy	Full GTM: channels, messaging, metrics, launch plan
beachhead-segment	First market segment identification
ideal-customer-profile	ICP with demographics, JTBD, needs
growth-loops	Sustainable growth flywheels
gtm-motions	PLG vs sales-led vs hybrid evaluation
competitive-battlecard	Sales-ready competitor comparison
marketing-ideas	Creative, cost-effective marketing ideas
positioning-ideas	Differentiated positioning from competitors
value-prop-statements	Value props for sales, marketing, onboarding
product-name	Product naming aligned to brand
competitor-analysis	Competitor strengths/weaknesses/differentiation
market-sizing	TAM, SAM, SOM estimation
pricing-strategy	Pricing, packaging, monetization
pptx	Slide deck file generation (.pptx)
growth-engine	Autonomous marketing experiments — run, measure, optimize (bootstrap CI, Mann-Whitney U)
sales-pipeline	Anonymous visitor → qualified pipeline (RB2B router, deal resurrector, ICP learner)
outbound-engine	ICP → automated cold outbound sequences, competitive monitor
seo-ops	Competitor keyword gaps, content attack briefs, GSC optimizer, trend scout
conversion-ops	Landing page CRO audit, survey-to-lead-magnet engine
yt-competitive-analysis	YouTube outlier videos and title-pattern extraction across competitor channels
x-longform-post	Human-sounding X/Twitter long-form posts + AI slop detector
podcast-ops	One podcast episode → 20+ content pieces across platforms
autoresearch	Content variant generation + expert-panel scoring + evolution loop

User Researcher

Skill	Purpose
user-personas	Refined personas from research data
market-segments	3–5 segments with demographics + JTBD
user-segmentation	Behavior/JTBD-based segmentation
customer-journey-map	End-to-end journey with touchpoints + pain points
interview-script	Structured interview scripts (Mom Test principles)
summarize-interview	Transcript → JTBD, satisfaction signals, actions
sentiment-analysis	Feedback sentiment + theme extraction