Guild PM Agent

A Guild.ai PM Agent that simulates the downstream consequences of roadmap sequencing decisions — built as part of the Guild application.

For James

npx guildaipmagent

Enter the access code when prompted. That's the only setup required.

The agent loads Guild.ai's GitHub backlog, generates 4 strategic sequencing scenarios, simulates the business consequences of each — which enterprise deals close or stall, when revenue unlocks, what risks emerge — and delivers a ranked recommendation.

What it does

1. Cinematic intro — sets the scene: you're the PM, 13 minutes to all-hands, three deals hanging
2. GitHub issue pull — streams 15 open issues from guildai/platform as if fetched live
3. Dependency analysis — deterministic graph validation, no LLM, same result every run
4. Scenario generation — 4 strategic archetypes: Enterprise First, Hub Launch Sprint, Revenue Optimized, Risk Minimized
5. Consequence simulation — Claude returns structured business impact per scenario: deals at risk, revenue timing, risk events
6. Recommendation — winning scenario with reasoning grounded in Guild.ai's current GTM reality

Architecture

Inspired by "You Can't Run Production on Probability".

guild-pm-agent/
├── cli/                    # npm package
│   └── src/
│       ├── index.ts        # Orchestrator
│       ├── cinema.ts       # Animations and story
│       ├── dependency.ts   # Deterministic dep-graph validator
│       ├── scenarios.ts    # Strategic archetype generator
│       ├── reporter.ts     # Terminal output + markdown report
│       └── data/
│           └── issues.json # 15 mocked Guild.ai GitHub issues
├── worker/                 # Cloudflare Worker
│   └── src/index.ts        # Bearer auth → parallel Anthropic calls → structured JSON
└── guild-agent.ts          # Native Guild SDK stub

guild-agent.ts

The guild-agent.ts at the repo root shows how this agent would be published natively to Guild Agent Hub using @guildai/agents-sdk — AutomaticallyManagedStateAgent with Zod input/output schemas, task.llm.generateText() for the narrative layer, and userInterfaceTools for progress notifications.

When Agent Hub is publicly available:

guild agent init --name guild-pm-agent --template AUTO_MANAGED_STATE
guild agent save --message "v1.0" --wait --publish

Built by Renan Serrano — Application to Guild.ai.