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.
minions 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 /minions), 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.
| 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 |
Each agent has a curated set of skills it invokes for specific tasks.
| 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 |
| 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) |
| 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 |
| 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 |
| 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 |
| Skill | Purpose |
|---|---|
grammar-check |
Grammar, logic, and flow checking for help articles |
emails |
Email sequences, drip campaigns, lifecycle emails |
onboarding |
Post-signup activation optimization |
churn-prevention |
Cancel flows, save offers, dunning |
content-ops |
Content quality scoring, editorial pipeline |
| Skill | Purpose |
|---|---|
test-scenarios |
Happy paths, edge cases, error handling |
webapp-testing |
Web application testing with Playwright |
| Skill | Purpose |
|---|---|
design-taste-frontend |
Generate new UI from scratch |
redesign-existing-projects |
Audit and improve an existing UI |
image-to-code |
Screenshot/mock → implemented UI code |
high-end-visual-design |
"Polished, calm, expensive" visual style |
minimalist-ui |
Notion/Linear-style restrained design |
industrial-brutalist-ui |
Industrial/Swiss-typography style |
stitch-design-taste |
Google Stitch-compatible design rules |
full-output-enforcement |
Prevents truncated generation output |
- Claude Code installed
- lark-cli installed and authenticated (
lark-cli auth login) - Node.js 18+
git clone https://github.com/bahni-m/minions.git
cd minions
./install.shThe script copies all agent files, installs the PM Ops skill, creates your config template, and installs all required skill packages automatically.
Open ~/.claude/skills/minions/config.md (created by the installer) and fill in your Lark base token and table IDs.
Open Claude Code in any workspace and type:
/minions
PM Ops will take your request, route it to the right subagents, and return a consolidated response.
You
│
▼ /minions
PM Ops ──────────────────────────────────────────────┐
│ │
├─── Associate PM (PRDs, OKRs, roadmaps, decks) │
├─── Data Analyst (SQL, KPIs, A/B, PostHog) │ parallel
├─── GTM Specialist (GTM, ICP, decks, competitive) │ or
├─── User Researcher (personas, journeys, research) │ sequential
├─── Customer Service (help center, emails, triage) │
├─── QA (test scenarios, QA reports) │
└─── UI/UX Designer (wireframes, design specs, UI) ─┘
PM Ops decomposes your request, dispatches independent subtasks in parallel and dependent ones sequentially, then aggregates everything into one response.
Just tell /minions what you need:
"Review sprint progress, reprioritize, create report, send to Lark, create demo deck, changelog, update wiki"
→ Data Analyst (velocity + blockers) → QA (readiness check) → Associate PM (report + deck + changelog + wiki) → PM Ops (sends to Lark)
"Prepare a .pptx pitch deck for our target customer segments"
→ User Researcher (extract JTBD from archetypes) → GTM Specialist (build slides) + Associate PM (validate claims)
"Setup North Star metric and KPIs for each pillar"
→ Data Analyst (define metrics) + Associate PM (map to Lark Base structure)
"Generate help center articles from PRDs, create triage issues, monitor feedback"
→ Customer Service (articles + triage + alerts)
"Write test scenarios for the new feature"
→ QA (happy path + edge cases + acceptance criteria check)
"Draft onboarding email sequence for new signups"
→ Customer Service (emails + onboarding skills, Day 0/3/7/14/30 with trigger conditions)
minions/
├── agents/
│ ├── minions-associate-pm.md ← Associate PM subagent
│ ├── minions-data-analyst.md ← Data Analyst subagent
│ ├── minions-gtm-specialist.md ← GTM Specialist subagent
│ ├── minions-user-researcher.md ← User Researcher subagent
│ ├── minions-customer-service.md ← Customer Service subagent
│ ├── minions-qa.md ← QA subagent
│ └── minions-ui-ux-designer.md ← UI/UX Designer subagent
├── skills/
│ ├── minions/
│ │ ├── SKILL.md ← PM Ops (entry point via /minions)
│ │ └── config.example.md ← Lark config template (copy → config.md and fill in)
│ └── optimize-goal/
│ ├── SKILL.md ← Review/enhance a goal.md task spec
│ └── references/
│ └── best-practices.md ← Prompt-engineering checklist used for the review
├── install.sh ← One-command installer
├── spec.md ← Full specification
└── README.md
After install, all files live in ~/.claude/agents/, ~/.claude/skills/minions/, and
~/.claude/skills/optimize-goal/.
| Tool | Purpose |
|---|---|
lark-cli base +... |
Read/write Lark Base (Issues, Pillars, Modules, Roadmap) |
lark-cli docs +... |
Read/write Lark Docs (PRDs, OKRs, wiki, changelogs) |
lark-cli im +... |
Send Lark messages (feedback alerts) |
| PostHog REST API | Product analytics — agents self-discover credentials from .env |
Lark base token and table IDs are stored in ~/.claude/skills/minions/config.md (filled in during setup). PostHog credentials, Lark group chat IDs, and Lark Doc tokens are self-discovered by agents at runtime.
Two remaining open questions before the team is fully operational:
| Item | How to resolve |
|---|---|
| Email engine | Tell Customer Service which engine you use (SendGrid, Mailchimp, Customer.io) for correct scheduling spec format |
| PostHog dashboards | Tell Data Analyst whether dashboards are greenfield or existing, to scope setup correctly |
Everything else self-configures on first use.
Found a way to improve an agent or skill definition? Have a new use case to add? Open an issue or PR — contributions to agent files, skills, and the install script are welcome.
See spec.md for the complete design including subagent definitions, orchestration protocol, use case walkthroughs, Lark Base schema, and success metrics.