| name | mlb-daily-probabilities |
| description | Regenerates today's MLB picks HTML and deploys to Cloudflare Workers. Does NOT rebuild wrangler.jsonc, src/worker.js, schema.sql, or any migrations — those are inputs and must exist before running. |
You are generating only MLB_Probabilities/index.html and MLB_Probabilities/props.html for today's MLB slate and deploying to Cloudflare Workers. You must not touch wrangler.jsonc, src/worker.js, schema.sql, or any migration file. If any required input is missing, abort and report — do not scaffold or improvise.
Background on why this skill is cautious: on 2026-04-22 a prior run regenerated wrangler.jsonc from scratch with a minimal assets-only config, which stripped the Worker's main, D1 binding, and secret. Every subsequent deploy wiped /api/*, tracking, leaderboard, and DOE instrumentation until a manual restore on 2026-04-24. Follow the rules below to the letter.
STEP 0 — LOCATE INPUTS (ABORT IF MISSING)
The scheduled task harness places project files under /sessions/*/mnt/uploads/. Required inputs, checked in this order:
uploads/wrangler.jsonc — Worker deploy config (main + D1 + assets binding + assets.run_worker_first: ["/api/*"])uploads/src/worker.js — Worker source (~477 lines, handles /api/*)uploads/MLB_Probabilities/ — static asset directory containing index.html, props.html, leaderboard.html, me.html, and any other files
If any of those is missing, produce a short report that lists exactly what's missing and stop. Do not generate placeholder config. Do not deploy.
Copy all inputs into a writable working directory (/sessions/*/mnt/outputs/work/) and do all subsequent edits there. Do not modify the originals in uploads/.
Verify:
grep -q '"main"' work/wrangler.jsonc || abort "wrangler.jsonc missing main field"
grep -q '"d1_databases"' work/wrangler.jsonc || abort "wrangler.jsonc missing d1_databases"
grep -q '"run_worker_first"' work/wrangler.jsonc || abort "wrangler.jsonc missing run_worker_first routing"
test -s work/src/worker.js || abort "src/worker.js missing or empty"
If any check fails, report and stop.
STEP 1 — RESEARCH TODAY'S SLATE
- MLB schedule, probable pitchers, late scratches
- Injury reports (IL moves, day-to-day questionables that affect lineups)
- Weather for outdoor parks (wind direction/speed, temp, precip)
- Consensus lines — FanDuel as anchor (moneylines, totals, F5, props: HR, hits, RBI, TB, SB, Ks)
STEP 2 — BUILD PROBABILITIES
De-vig the FanDuel anchor to get implied probability. Adjust ±3–6pp based on:
- Pitcher handedness vs batter splits (L/R, wOBA, ISO)
- Recent form (last 15 games)
- Weather multipliers (wind out = +HR, cold/wind in = -HR)
- Park factors
- Bullpen fatigue / pen-burn
Pick edges ≥3pp over de-vigged line. Grade confidence:
- High: ≥6pp edge, strong multi-factor support
- Medium: 3–6pp edge, 2+ factors
- Low: edge present but fragile / variance-heavy
STEP 3 — REGENERATE work/MLB_Probabilities/index.html AND work/MLB_Probabilities/props.html ONLY
Preserve everything else in work/MLB_Probabilities/ exactly as-is. Do NOT touch leaderboard.html, me.html, .assetsignore, or any other file.
Use the existing HTML in uploads/MLB_Probabilities/ as the structural template — preserve the CSS block, nav, theme, tooltip CSS+JS, updated-badge CSS, and any JS that hooks up Track buttons. Only the data content changes.
STEP 3a — PICK PILL ATTRIBUTES (MANDATORY — THE TRACKING API DEPENDS ON THESE)
Every pick pill (<div class="pill ...">) must carry:
data-pick-id="{YYYY-MM-DD}:{type}:{game-slug}:{detail-slug}" where type ∈ ml, total, f5, prop, parlay. Example: 2026-04-24:prop:HOU-NYY:yordan-alvarez:over-1-5-tbdata-conf-score="{0-100}" — integer model confidence (High tier 70-90, Med 55-70, Low 45-55)data-edge-pct="{decimal}" — model probability minus de-vigged market probability, as decimal (e.g. 0.065 for +6.5pp)data-weather-cert="{0-1}" — confidence in weather data (1.0 for domes/predictable, 0.6 for unsettled)data-lineup-cert="{0-1}" — confidence that key batters/pitchers will play (1.0 confirmed lineup, 0.7 probable, 0.4 questionable)data-factors='{"park_hr_factor":1.08,"splits_edge_pp":4.2,"form_l15":0.320,...}' — JSON blob of bet-type-specific factors. Escape quotes for HTML attribute.data-reason="<plain-English one-paragraph explanation>" — used by the existing tooltip CSS
Include data-tip-side="left" on right-column pills to avoid tooltip viewport overflow.
STEP 3b — STAKE PILL AND UNIT SIZING
Straight plays: 1.00u baseline, scale 0.5–1.5u by confidence.
Parlays (combined American odds):
- +150 to +250 → 0.50u
- +250 to +500 → 0.25u
- +500 to +1000 → 0.10u
- HR-pair parlays (2 HR legs) → 0.10u fixed
- +1000 to +2500 → 0.05u
-
+2500 → 0.02u
Every pick renders <span class="stake-pill">0.25u</span>. Include a unit-guide legend block on both pages.
STEP 3c — LAST-UPDATED BADGE
Top of each page: <span class="updated-badge"><span class="dot"></span> {Day}, {Mon DD, YYYY} · {HH:MM} AM ET</span> with the pulsing green dot CSS from the existing template. Use actual current day/date/time (ET).
STEP 3d — TEAM LOGOS
Logo URL pattern: https://www.mlbstatic.com/team-logos/{MLB_TEAM_ID}.svg
ARI=109 ATL=144 BAL=110 BOS=111 CHC=112 CWS=145 CIN=113 CLE=114
COL=115 DET=116 HOU=117 KC=118 LAA=108 LAD=119 MIA=146 MIL=158
MIN=142 NYM=121 NYY=147 OAK=133 (ATH=133) PHI=143 PIT=134 SD=135
SF=137 SEA=136 STL=138 TB=139 TEX=140 TOR=141 WSH=120
props.html: inside each <div class="player">, prepend <img class="team-logo" src="https://www.mlbstatic.com/team-logos/{ID}.svg" alt="{ABBREV}" loading="lazy" /> using the card's data-team abbrev.
index.html: each <div class="matchup">Team A @ Team B</div> renders as logo+name @ logo+name using .team-logo.sm.
STEP 4 — DEPLOY
Do not regenerate wrangler.jsonc. Do not add/remove bindings. Run:
cd work
CLOUDFLARE_API_TOKEN="$CF_TOKEN" CLOUDFLARE_ACCOUNT_ID="9393ba26563604d847b6e1d03a2faa55" \
npx --yes wrangler@4.83.0 deploy
Wrangler will read main, d1_databases, assets from the existing wrangler.jsonc. If the deploy output does NOT list env.DB (mlb-tracking) and env.ASSETS, abort and report — the config was tampered with.
The API token lives in $CF_TOKEN (scheduled task environment variable). If unset, abort — do not hardcode a token in this file.
STEP 5 — POST-DEPLOY VERIFICATION
Immediately after deploy, query the active version via API and confirm it has at least assets:ASSETS, d1:DB, and secret_text:SETTLE_SECRET:
curl -s "https://api.cloudflare.com/client/v4/accounts/9393ba26563604d847b6e1d03a2faa55/workers/scripts/mlb-probabilities/deployments?order=desc&per_page=1" \
-H "Authorization: Bearer $CF_TOKEN" | jq -r '.result.deployments[0].versions[0].version_id'
If fewer than 3 bindings are present, wrangler rollback $PREV_VERSION and report. Do not leave a broken deploy live.
STEP 6 — HEALTH CHECKS (rollback on fail)
Fetch live site (expect 302 to Cloudflare Access login — that's fine, the Worker is up). Then fetch the asset-serve-directly mode via a known-good probe. Verify in the generated HTML files in work/MLB_Probabilities/:
- Today's date string appears in badge on both pages
- Pulsing green dot CSS present (
updated-badge, .dot)
- ≥20
data-reason= occurrences across both pages
- ≥20
data-pick-id= occurrences across both pages
- ≥20
data-conf-score= occurrences across both pages (new DOE requirement)
- ≥20
data-edge-pct= occurrences across both pages
- ≥5
stake-pill occurrences
unit-guide block on both pages- ≥10
class="team-logo occurrences on each page
- ≥10 tooltips (
data-reason) on props.html
If ANY check fails: wrangler rollback $PREV_VERSION and report.
STEP 7 — REPORT
Brief 3–5 line summary: slate size, top-confidence picks, total units staked, deploy version ID, any rollback/issue. Include live URL https://mlb-probabilities.bbaker-939.workers.dev/.
THINGS THIS SKILL MUST NEVER DO
- Regenerate
wrangler.jsonc from scratch
- Regenerate
src/worker.js
- Strip bindings from
wrangler.jsonc
- Change
compatibility_date
- Scaffold a minimal config if files are missing — report instead
- Hardcode a Cloudflare API token in this file
- Deploy if post-deploy binding verification fails
