`pm init`

Product Introduction

1. Definition: pm init is the entry point for PM, an open-source Product Manager CLI (Command Line Interface) designed to perform deterministic, automated audits of software products. Technically classified as a Product Management Audit Tool and Developer Experience (DX) utility, it resides in the local development environment and leverages Large Language Models (LLMs)—specifically Claude 3.5 Sonnet—to analyze the codebase as the primary source of truth for product behavior and user experience.

2. Core Value Proposition: The tool exists to resolve the "Intent-Reality Gap" in software development. By cross-referencing the actual source code (routes, forms, components) with optional marketing positioning, pm init identifies where the builder’s goals diverge from the user’s actual journey. It automates high-leage product strategy by generating actionable product specifications and working HTML prototypes, allowing technical founders and product owners to find and fix friction points without manual session recording analysis or subjective user interviews.

Main Features

1. Deterministic Six-Stage Analysis Pipeline: Unlike generic AI chat interfaces, PM operates through a structured, reproducible sequence of events. The pipeline begins with Intent Extraction (reconstructing a Lean Canvas from code), moves to Imagined Critical Path mapping (inferring user flow from routes), executes Delta Analysis (naming intent mismatches), performs Leverage Prioritization (ranking problems based on impact), generates Solution Specs + Prototypes (working HTML and Markdown), and concludes with Artifact Assembly.

2. Static Codebase Contextualization: PM reads the local repository to extract the "source of truth." It identifies technical implementation details such as authentication gates, tracking events, and UI components to narrate the user experience directly from the logic. Unlike dynamic testing tools, it does not require running the application or bypassing auth-gated screens, making it compatible with complex production apps, server-side infrastructure, and seed-data-dependent environments.

3. Marketing-to-Code Cross-Referencing: During the audit process, the tool can optionally fetch a public marketing URL (landing pages, pricing, feature lists). This feature uses the marketing copy as a secondary data point to verify if the product's positioning promise matches the technical reality. If the code contradicts the marketing (e.g., a promised feature is unreachable or a signup flow is broken), the CLI reports the divergence as a critical finding.

4. Local-First Privacy and Security: Built with a security-conscious architecture, PM stores the user’s Anthropic API key locally in ~/.pm/config.yaml with restricted permissions (chmod 600). No data is transmitted to a PM-specific backend; all inference calls are made directly to api.anthropic.com. This ensures that sensitive proprietary code remains between the user’s machine and the LLM provider.

Problems Solved

1. Pain Point: Abandoned User Journeys and UX Debt. Many products suffer from "zombie features"—code that exists but is inaccessible—or signup flows that lead to high bounce rates due to technical friction. PM identifies these "intent mismatches" where the code fails to deliver the promised user value.

2. Target Audience: The tool is specifically designed for Product Owners, Solo Founders, Technical Product Managers, and Full-stack Developers who manage complex repositories and require objective, data-driven feedback on their product's UX without the overhead of traditional user testing.

3. Use Cases: PM is essential for pre-launch audits to ensure the MVP matches the marketing site; post-update "sanity checks" to see if new features have cluttered the critical path; and technical debt assessments where a product manager needs to justify refactoring based on user journey leverage rather than just code cleanliness.

Unique Advantages

1. Differentiation: Traditional product management tools (like Jira or Mixpanel) are either management-heavy or data-heavy, requiring human interpretation of events. PM is an "agentic" tool that performs the interpretation itself by reading the code. It differs from user testing platforms because it requires zero real-world users to generate insights; it finds flaws in the logic of the product itself.

2. Key Innovation: The use of an LLM as a deterministic auditor rather than a creative assistant. By constraining the AI to a six-stage YAML-producing pipeline, PM minimizes hallucinations and ensures that the output (the spec.md and spec.html) is structured, high-leverage, and grounded in the actual routes and components found in the repository.

Frequently Asked Questions (FAQ)

1. How much does it cost to run a product audit with pm init? A typical audit using Claude 3.5 Sonnet takes approximately 8–12 minutes and costs between $1 and $3 in Anthropic API credits, depending on the size of the codebase and your specific tier pricing. Deep-tier audits with more extensive analysis may cost 3–5 times more.

2. Does PM need access to my production environment or database? No. PM never runs your product, never signs in as a user, and never interacts with your production database or auth-gated screens. It performs its analysis entirely by reading your source code locally and optionally fetching your public marketing website.

3. Can I use PM with other LLMs like GPT-4 or local models? By default, PM uses Anthropic's Claude 3.5 Sonnet for inference. However, since the project is open-source (MIT License), users can modify the source code to support other models or contribute via a Pull Request on GitHub to add official support for different LLM providers.