draft-prep - SKILL.md Agent Skill

name: draft-prep description: Run the pre-season fantasy football draft preparation workflow. Generates preseason projections from historical data, fetches live ADP from Sleeper or FantasyPros, computes value scores, and launches the interactive draft assistant. Use in the weeks before an NFL draft (typically August). argument-hint: "[season] [scoring] [my-pick] [teams]" allowed-tools: Bash, Read, mcpsleeper*, mcpfetchfetch

Run the complete pre-season fantasy football draft preparation workflow.

Arguments

$ARGUMENTS — parsed as: [season] [scoring] [my-pick] [teams]

Defaults: season=2026, scoring=half_ppr, my-pick=1, teams=12
Scoring options: ppr, half_ppr, standard

Current projection files

!cd /Users/georgesmith/repos/nfl_data_engineering && ls output/projections/ 2>/dev/null | grep -i preseason | tail -5 || echo "No preseason projections yet"

Steps

Step 1 — Check for existing preseason projections

If a recent preseason projections CSV exists in output/projections/, offer to use it or regenerate.

Step 2 — Generate preseason projections (if needed)

source venv/bin/activate && python scripts/generate_projections.py \
  --preseason \
  --season SEASON \
  --scoring SCORING \
  --output both

This fetches 2 seasons of historical data from nfl-data-py and produces ranked projections.

Step 3 — Fetch ADP data (use MCP, then save to data/adp.csv)

Option A — Sleeper MCP (preferred, always current): Use the sleeper MCP to fetch current ADP rankings. Call the Sleeper trending/ADP endpoints to get player draft positions, then save to data/adp.csv with columns player_name,adp_rank.

Option B — fetch MCP (FantasyPros fallback): Use the fetch MCP to retrieve the FantasyPros consensus ADP page and parse player rankings:

URL: https://www.fantasypros.com/nfl/adp/overall.php
Extract player name and ADP rank from the rankings table
Save to data/adp.csv

After saving ADP data, confirm row count and show the top 10 ADP players.

Step 4 — Show pre-draft summary

After projections and ADP are loaded, display:

Top 30 overall players by model rank vs ADP rank
Top 5 per position (QB/RB/WR/TE)
Top 10 undervalued players (model rank >> ADP rank)
Top 5 overvalued players to avoid (ADP rank >> model rank)
Positional depth: how many viable starters at each position

Step 5 — Launch interactive draft assistant

source venv/bin/activate && python scripts/draft_assistant.py \
  --scoring SCORING \
  --teams TEAMS \
  --my-pick MY_PICK \
  --season SEASON \
  --projections-file output/projections/LATEST_PROJECTION_FILE \
  --adp-file data/adp.csv

Draft assistant quick-reference

Once the draft starts, remind the user of key commands:

rec — get pick recommendations for your turn
pick <name> — you draft this player
draft <name> — another team drafted this player
best [QB/RB/WR/TE] — best available by position
undervalued — players the model likes more than ADP
roster — see your current roster
undo — undo your last pick

Notes

Sleeper MCP gives real-time ADP; re-run Step 3 on draft day for the most current data
Best run 1–2 weeks before draft day so projections reflect final depth charts
Re-run after major injuries or trades to refresh rankings
Save the draft results for season-long reference