PM Agent Extension — AI-powered Product Management for OSS projects

[ASekiguchi]

What would you like to be added?

A Gemini CLI Extension that provides AI-powered Product Management capabilities as a reusable, bottom-up framework for OSS projects. This extension would analyze existing project artifacts (Issues, PRs, commits, docs, roadmap) and generate structured PM outputs — without requiring a human Product Manager.

Why is this needed?

The structural gap in OSS product management

Most OSS projects, including Gemini CLI, are engineering-led — which is a strength for code quality but creates a structural gap in product management:

Function	Current state in typical OSS	Impact
Requirements gathering	Ad-hoc via Issues and discussions	Fragmented, no single source of truth
Prioritization	Maintainer intuition + community pressure	Inconsistent, not data-driven
Roadmap management	Static markdown file (ROADMAP.md)	Disconnects from actual development activity
User feedback synthesis	Manual reading of Issues	Time-consuming, insights get lost
Release planning	Engineering-driven milestone tracking	Lacks user-impact perspective
Competitive analysis	Not systematically tracked	Reactive instead of proactive

Why a bottom-up AI approach is better than hiring a PM

Traditional PM involvement in OSS has friction:

• Cultural mismatch: PMs use different tools (Jira, Notion, Confluence) than developers (GitHub, terminal)
• Context switching cost: Developers must context-switch to PM workflows
• Scalability: Human PMs can't be embedded in every OSS project

A bottom-up AI PM Agent solves this by:

1. Working where developers already work — inside the CLI and GitHub
2. Deriving insights from existing artifacts — no new data entry required
3. Producing developer-friendly outputs — markdown, not slide decks
4. Scaling across projects — one extension, any OSS project

Enterprise adoption accelerator

This directly addresses a key adoption barrier: enterprises evaluating Gemini CLI need structured product artifacts (PRDs, roadmaps, risk assessments) that OSS projects typically don't maintain. An AI PM Agent can generate these on-demand, bridging the gap between OSS development culture and enterprise procurement requirements.

Proposed design

Architecture: Gemini CLI Extension

Using Gemini CLI's existing extension framework (no core fork required):

pm-agent/
├── gemini-extension.json        # Extension manifest
├── GEMINI.md                    # PM context & principles
├── mcp-server/
│   ├── server.js                # MCP server with PM tools (Node ESM, no build step)
│   └── server.test.js           # Unit tests (node:test)
├── skills/
│   ├── requirements-analyzer/SKILL.md
│   ├── roadmap-generator/SKILL.md
│   └── backlog-prioritizer/SKILL.md
├── commands/pm/
│   ├── analyze.toml             # /pm:analyze
│   ├── health.toml              # /pm:health
│   ├── roadmap.toml             # /pm:roadmap
│   └── prioritize.toml         # /pm:prioritize
└── references/frameworks/
    ├── swot.md
    ├── moscow.md
    └── rice-scoring.md

MCP Tools

• analyze_project_issues — Fetches and categorizes GitHub Issues (feature requests, bugs, UX feedback, tech debt). Label matching uses normalized regex patterns that handle conventions like kind/, type:, area/.
• analyze_commit_velocity — Tracks development velocity via GitHub Commits API (not local git).
• check_project_health — Aggregates issue/PR counts via GitHub Search API (no pagination cap) and velocity trends.

Skills

• requirements-analyzer — Extracts user stories and acceptance criteria from Issues/PRs
• roadmap-generator — Creates Now/Next/Later roadmaps from issue and commit activity
• backlog-prioritizer — Applies RICE and MoSCoW frameworks to open issues

Example interactions

> /pm:analyze
→ Generates SWOT, top user pain points, RICE-scored priorities

> /pm:health
→ Shows velocity trends, issue resolution rates, risk flags

> /pm:prioritize
→ Applies RICE scoring to open Issues, suggests sprint planning

> /pm:roadmap
→ Generates a Now/Next/Later roadmap from current activity

> "What are the top user pain points right now?"
→ The requirements-analyzer skill activates automatically

Why this should be an OSS extension (not internal tool)

1. Every OSS project needs PM — This extension is project-agnostic. Any repo with Issues and a README can benefit.
2. Dog-fooding Gemini CLI's extension system — Demonstrates the power of the extension architecture to the community.
3. Attracts non-engineer contributors — Connects to the non-engineer contribution proposal (Enable non-engineer contributions (PM, UX Designer) to the OSS project #20495). PM Agent outputs create entry points for non-engineers to engage.
4. Enterprise value — Companies evaluating OSS tools can run /pm:analyze to get instant product intelligence.

Implementation phases

Phase 1: Core Analysis (MVP) — implemented

• MCP server with analyze_project_issues, analyze_commit_velocity, check_project_health
• Skills: requirements-analyzer, roadmap-generator, backlog-prioritizer
• Commands: /pm:analyze, /pm:health, /pm:roadmap, /pm:prioritize
• Unit tests (46 passing), Apache-2.0 license, no build step, Node ESM
• Works with any authenticated GitHub repository

Phase 2: Planning Tools (future)

• RICE/MoSCoW scoring enhancements, competitive analysis skill, /pm:prd generation

Phase 3: Automation & Integration (future)

• Hooks for automatic context injection, release planning integration, scheduled health reports

Technical feasibility

This proposal uses only existing Gemini CLI extension capabilities:

• MCP servers for tools (GA)
• Skills for specialized workflows (GA)
• Custom commands for user entry points (GA)
• No core modifications required
• No fork needed

Additional context

This proposal builds on Issue #20495 (Enable non-engineer contributions) and addresses the broader challenge of making OSS projects more product-aware. The AI PM Agent is a concrete mechanism to bridge the gap between engineering-led development and product-informed decision making.

The approach is bottom-up by design: instead of imposing PM processes on developers, it derives PM insights from artifacts developers already create (Issues, PRs, commits, docs). This makes it developer-friendly and zero-friction to adopt.

---

Update (2026-04-10)

Since this issue was filed, the Gemini CLI extension system has matured significantly. MCP servers, Skills, and custom commands are now well-documented and stable. This strengthens the feasibility of the proposal: the entire PM Agent can be built as a pure extension contribution with no core modifications required.

Revised framing:

This is best understood as a high-value ecosystem extension — a showcase of what the Gemini CLI extension platform can do for non-engineering workflows. The primary justification is simple: GitHub-based OSS projects generate rich structured data (Issues, PRs, commits) that today goes unanalyzed from a product perspective. A well-designed extension can surface that data as actionable PM insights with no new tooling or process changes required.

On the ROADMAP.md rationale:

The original proposal included comparing static ROADMAP.md files against actual activity. In practice, most active OSS projects — including Gemini CLI — now track priorities through GitHub Issues and milestones rather than a static roadmap file. The roadmap-alignment use case is still valid but should be treated as secondary. The stronger MVP focus is on Issue-based analysis: categorization, RICE scoring, and health metrics derived from live GitHub data.

Revised MVP scope (Phase 1 only):

A working implementation already exists at extensions/pm-agent/ in the contributor's fork. It includes:

• MCP server with three tools: analyze_project_issues, analyze_commit_velocity, check_project_health
• Three Skills: requirements-analyzer, roadmap-generator, backlog-prioritizer
• Four commands: /pm:analyze, /pm:health, /pm:roadmap, /pm:prioritize
• Unit tests (46 passing), Apache-2.0 license, no build step, Node ESM

All GitHub data is fetched via the gh CLI using execFile (no shell injection). Counts use the GitHub Search API to avoid pagination caps. The extension works against any authenticated GitHub repository.

Phases 2 and 3 from the original proposal (automation, hooks, PRD generation) remain valid future work but are explicitly out of scope for the initial PR.

[ASekiguchi]

I would like to implement this extension. Specifically, I am planning to create:

1. Extension scaffold (gemini-extension.json, GEMINI.md) — configuration only
2. Skills (SKILL.md files) — markdown instructions for PM workflows (requirements analysis, roadmap generation, backlog prioritization)
3. Custom commands (TOML files) — /pm:analyze, /pm:health, /pm:roadmap, /pm:prioritize
4. MCP server (JavaScript) — tools for GitHub issue analysis and commit velocity tracking, following the existing MCP example pattern

This builds on Issue #20495 (non-engineer contributions) and would serve as a practical example of an extension that bridges product management and engineering workflows.

Could a maintainer review this proposal and apply the help-wanted label if appropriate? I am happy to start with a minimal MVP (skills + commands only) and iterate based on feedback.

/assign

[gemini-cli]

👋 @ASekiguchi, thanks for your interest in this issue! We're reserving self-assignment for issues that have been marked with the help wanted label. Feel free to check out our list of issues that need attention.

[ASekiguchi]

Update: Working prototype built

I've built a working prototype of the PM Agent Extension using only existing extension capabilities (no core modifications).

Architecture

Follows the patterns in packages/cli/src/commands/extensions/examples/:

• MCP Server (JS) — 3 tools: analyze_project_issues, analyze_commit_velocity, check_project_health
• Skills (SKILL.md) — requirements-analyzer, roadmap-generator, backlog-prioritizer
• Commands (TOML) — /pm:analyze, /pm:health, /pm:roadmap, /pm:prioritize
• Reference frameworks — RICE, MoSCoW, SWOT (markdown)

Test results (run against google-gemini/gemini-cli)

Tool	Result
analyze_commit_velocity	1,411 commits / 3 months, 108 contributors, velocity +7.8% (stable)
analyze_project_issues	Fetches and auto-categorizes issues (feature / bug / UX / tech debt)
check_project_health	Open issues: 30, Closed (30d): 100, PRs merged (30d): 100, 107 active contributors

All 3 MCP tools pass with real data. Skills and commands are markdown/TOML (no TypeScript).

File structure (15 files total)

extensions/pm-agent/
├── gemini-extension.json
├── GEMINI.md
├── package.json
├── mcp-server/server.js
├── skills/{requirements-analyzer,roadmap-generator,backlog-prioritizer}/SKILL.md
├── commands/pm/{analyze,health,roadmap,prioritize}.toml
└── references/frameworks/{swot,moscow,rice-scoring}.md

Happy to adjust scope or approach based on feedback. A minimal PR could start with just skills + commands (no MCP server) if preferred.

[cocosheng-g]

Triage Summary

This issue is a comprehensive proposal for a PM Agent Extension designed to bring AI-powered Product Management capabilities (requirements analysis, roadmap generation, backlog prioritization) to OSS projects. A community member has already developed a working prototype and shared detailed architecture and test results. Since this is a well-defined feature request that enhances the Gemini CLI extension ecosystem and is ready for external contribution, it is being categorized as Help-wanted.

[ASekiguchi]

@cocosheng-g
I would like to work on this issue.
Could you please assign it to me？

[ASekiguchi]

I've refreshed the issue description to reflect the current state of the Gemini CLI extension system and to clarify the MVP scope.

The core proposal is unchanged: an extension that surfaces PM insights (issue categorization, velocity trends, backlog prioritization) from existing GitHub artifacts with no core modifications required.

Two small framing updates:

1. Extension maturity: The extension system (MCP servers, Skills, custom commands) is now well-documented and stable, which makes this a cleaner contribution than when the issue was first filed.

2. Roadmap framing: The original mention of comparing static ROADMAP.md files against activity has been softened. Most active projects track priorities through GitHub Issues and milestones now, so the stronger justification is Issue-based analysis.

A working implementation (extensions/pm-agent/) is ready for review. It covers the Phase 1 MVP: three MCP tools, three Skills, four commands, 46 unit tests, no build step. Happy to open a PR when the maintainers confirm the direction.

[cocosheng-g]

Sorry for the confusion. Please follow https://geminicli.com/docs/extensions/releasing/ to create and publish your extension.