Skip to content
v0.9.0CC-BY-SA-4.0

AI-First Product Manager skills: synthetic personas, persona interviews (exploration and validation), insight extraction, persona critique panels, and exposure plans.

afpm — AI-First Product Manager

Agent skills for discovery and validation with synthetic users. Companion plugin for the AI-First Product Manager program by Alaimo Labs.

Overview

This plugin gives your coding agent a product-discovery toolkit: frame opportunities before choosing solutions, create synthetic personas, interview them, extract insights, run critique panels over your specs, and slice features into exposure plans. It also bridges to real research: design interview guides and surveys, analyze the results, and derive evidence-based personas from the patterns. All artifacts are plain markdown files in your repo under product/ — no external services.

Content is written in English; all deliverables come out in the language you work in.

Install

From the ai-first-skills marketplace in Claude Code:

/plugin install afpm

The plugin also ships a portable Agent Plugins manifest, so compatible clients (Cursor, VS Code, GitHub Copilot, ChatGPT & Codex, Kiro, OpenClaw, Hermes) can load it too — see the repo README for per-client instructions.

Workflows

User-invoked skills — you trigger them as slash commands; they never auto-load.

WorkflowWhat it does
/start-productBootstrap product/overview.md: mode, context, sponsor (internal), ranked and tagged unverified beliefs
/frame-opportunityFrame a problem for a segment — signals, beliefs, research agenda — before any solution
/research-marketSecondary research/benchmarking, every claim provenance-tagged
/generate-personasGenerate a diverse set of synthetic personas — typed primary/secondary/tertiary/negative, mix on request or proposed
/interview-personaInterview a persona — exploration or validation mode
/extract-insightsExtract actionable insights from transcripts (synthetic/real)
/design-interviewDesign an interview guide + recruitment plan (survey opt-ins first) for real-user research
/test-interview-guidePretest a guide against a persona and fix what breaks
/design-surveyDesign a survey questionnaire with an interview opt-in block, ready for any survey tool
/analyze-surveyAnalyze survey results: quant summary, themes, insights
/derive-personasDerive evidence-based personas from real research patterns
/map-frictionsMap cognitive frictions across a journey's steps (MFC)
/clarify-ideaSharpen a fuzzy idea via one-question-at-a-time brainstorming; names its parent opportunity or declares none
/write-specDraft an evidence-grounded spec: journey, stories, criteria
/critique-specPersona panel critiques a spec/PRD, with synthesis
/slice-featureTurn a spec's hypothesis into an Exposure Plan
/review-evidenceWeekly sweep: new evidence vs. beliefs, drift report, corrections log

Knowledge skills

Model-invoked — the agent loads them automatically when the topic matches.

SkillKnowledge it carries
opportunity-framingOpportunity vs. solution vs. outcome, signals vs. proof, research agenda, the OST as files
secondary-researchProvenance discipline, source hierarchy, lanes, belief mapping
synthetic-personasArchetype principles, persona structure, the four persona types, diversity requirements
synthetic-interviewsIn-character interview roleplay; exploration vs. validation modes
insight-extractionFocus areas, grounding rules, insight quality bar
interview-guidesDiscussion-guide design: goals → questions, funnel, non-leading; pretesting
survey-designQuestionnaire craft, wording bias, scales, results analysis
cognitive-frictionsThe MFC lens: four friction categories, severity, opportunity bar
feature-specsSpec structure: journey, stories, criteria, hypothesis, risk-tagged assumptions
persona-critiqueIn-character document reviews and panel synthesis
exposure-plansBuild ≠ reveal, belief decomposition, level design, validations

File conventions

Artifacts live in your repo:

product/
├── overview.md          # product context, sponsor (internal), tagged + ranked belief registry
├── corrections.md       # log of human corrections to AI proposals (kept by /review-evidence)
├── personas/            # one file per persona (synthetic or derived), each with a `type:` — primary / secondary / tertiary / negative
├── interviews/          # transcripts, synthetic and real
├── interview-guides/    # guides for real-user interviews
├── surveys/             # survey questionnaires
├── insights/            # extracted insights, survey analyses & critique panels
├── research/            # secondary research & benchmarks
├── journeys/            # user journeys + cognitive friction maps
├── opportunities/       # opportunity briefs: problem + segment + signals + research agenda, no solution yet
├── ideas/               # clarified idea briefs (each names its parent opportunity, or `none (declared)`)
├── specs/               # feature specs
└── exposure-plans/      # exposure plans

overview.md is the single belief registry: every unverified belief lives there — never in per-opportunity, per-idea, or per-spec lists — tagged by scope ([product], [opportunity: {slug}], or [feature: {slug}]) and risk ([value] / [usability] / [feasibility] / [viability]), ranked by impact × uncertainty. Internal products also record their sponsor: who funds the product and what they need to see to keep funding it.

The three scopes mirror an Opportunity Solution Tree: overview.md (outcome + product beliefs) → opportunities/ (a problem for a segment, framed by /frame-opportunity) → ideas/ (candidate solutions, each hanging from an opportunity) → specs/exposure-plans/. The opportunity level is optional — you can take a decided feature straight to /clarify-idea — but skipping it is a declared decision: the idea brief records opportunity: none (declared) and registers the problem it assumes as a belief.

As evidence arrives, beliefs in overview.md get a status appended on the belief's own line — — confirmed/contradicted/weakened by [file] (date) (keywords stay in English, like source: values; no status = still unverified). Only evidence from real users confirms; synthetic evidence just makes a belief promising. /review-evidence, /extract-insights, and /analyze-survey propose these annotations — you approve before anything is written.

License

CC BY-SA 4.0 © Alaimo Labs. Use, adapt, and share freely — credit Alaimo Labs and keep derivatives under the same license.