Last August, I wrote about building an AI-powered fantasy football draft assistant by juggling multiple platforms: NotebookLM for research synthesis, ChatGPT for strategy, Claude for detailed analysis, and a spreadsheet to tie it all together. It worked, but it felt like coordination overhead. You had to context-switch between tools, manually copy findings between windows, and paste results into a doc. Not terrible for one draft per year, but not elegant either.
Today I'm releasing a Claude Code skill that brings that entire workflow into a single tool.
The Original Problem
Back then, the challenge was simple: fantasy football data is perishable. Your ADP spreadsheet from May is worthless by August. Expert consensus shifts weekly. Injuries happen. Offseason moves land. The only way to make smart draft decisions is to synthesize fresh data from multiple sources in real time.
So I built a workflow that pulled from FantasyPros, ESPN, PFF, and a dozen specialty sites. NotebookLM synthesized research notes. ChatGPT did strategy analysis. Claude handled depth. It worked great. It also took about three hours of active setup per draft season.
The skill automates all of that.
What the Skill Does
The fantasy-football skill has four modes. Each one follows the same principle: web research first (never trust training data for fantasy rankings), structured analysis, and actionable output.
Draft Prep: Spawns four parallel research agents scoped to rankings, injuries, matchups, and advanced analytics. Returns a comprehensive .docx with tiered rankings by position, round-by-round strategy, sleeper candidates, players to avoid, and playoff schedule advantages. This mirrors the old multi-AI workflow but runs in one invocation.
/fantasy-football draft
Live Draft: Interactive real-time assistant. You type player picks as they happen. When it's your turn, one focused agent does quick web research on current ADP and news, then gives you top-3 recommendations with one-line rationale. No depth analysis, because the clock is ticking.
/fantasy-football live-draft
Weekly Matchup: Scoped to a single week during the season. Four agents in parallel research projections, injuries, opponent matchups, and usage trends. Output is a .docx with start/sit recommendations, FLEX decisions, waiver wire targets, and what to watch next week.
/fantasy-football week 8
Trade Analyzer: Two agents evaluate the players involved. Terminal-only output, decision in under 30 seconds. Accept, reject, or counter recommendation.
/fantasy-football trade Gibbs and Puka for Henry and Kupp
Why This Matters
The skill demonstrates a pattern that works for any time-sensitive decision with perishable data: parallel research agents, structured cross-referencing, confidence scoring, and web-first methodology. I built this because fantasy football is my personal use case, but the same architecture applies to hiring decisions, investment research, competitive analysis, or any situation where you need fast synthesis of current information from multiple sources.
It also shows what happens when you actually test and iterate on a workflow before you try to automate it. I ran the old version dozens of times before I knew which agents I needed, what they should research, and how to structure the handoff between research and recommendation. That knowledge went straight into the skill.
The Implementation
The skill uses Claude Code's multi-agent orchestration under the hood. Each mode spawns agents with focused research briefs. Agents use WebSearch and WebFetch to pull current data (not training knowledge). Results get cross-referenced for confidence scoring and contradictions. Final output is either a .docx report or terminal recommendation.
Configuration is stored in a simple JSON file that you edit once (here’s a simplified example):
{ "league": { "teams": 10, "scoring": "PPR", "draft_type": "snake", "roster_size": 16, "starters": 9, "positions": { "QB": {"start": 1, "max": 4}, "RB": {"start": 2, "max": 8}, "WR": {"start": 2, "max": 8}, "TE": {"start": 1, "max": 3}, "FLEX": {"start": 1, "eligible": ["RB", "WR", "TE"]}, "DST": {"start": 1, "max": 3}, "K": {"start": 1, "max": 3} }, "bench": 7, "ir": 1 }, "draft_position": 5, "roster": {}, "season_year": 2026 }
Live draft mode tracks your picks as you make them. Weekly mode reads your roster from the config. Trade analyzer pulls context from what you already have. Everything else is orchestration and web research. The whole thing is open source so you can inspect, fork, or extend it however your league demands.
Open Source, Available Now
The skill is open source and available on GitHub. If you're using Claude Code, you can clone the repo into your .claude/skills/ directory and start using it immediately.
No sign-ups, no separate accounts, no training data workarounds. Just honest web research and structured multi-agent thinking.
What's Next
The evolution from the original August workflow to this skill taught me something worth iterating on: automation should follow validation. I didn't decide the four research agents in draft prep were the right set until I'd manually run that workflow a dozen times. I didn't know live draft needed speed optimization until I actually stood in front of the draft board. Those constraints became features of the skill.
If you use it and find patterns that don't work, file an issue on GitHub. If you want to extend it (weekly sit recommendations, league-specific analysis, integration with Sleeper or ESPN APIs), pull requests are welcome.
For now, the next draft is covered. No more juggling tabs.