InferSports Skill

Live Asian-priced football & basketball odds, scores, and de-vigged fair lines —
packaged as an Agent Skill for Claude Code,
Codex, OpenClaw, or any agent that can run a shell script.

Read-only · Keyless · 11 verbs · one line of output per answer

$ scripts/match.sh "Brazil vs Argentina" --tz America/Sao_Paulo
Brazil vs Argentina — LIVE 1h 12 1-0.
  favorite: Brazil (61.4%)

$ scripts/fair.sh "France vs Argentina"
France v Argentina — sharp fair (1x2): home 55.6% / draw 28.2% / away 22.5% · de-vig pinnacle

$ scripts/compare.sh "France vs Argentina" --prob 0.52 --outcome home --label polymarket
France v Argentina — 1x2 home: sharp fair 55.6% vs polymarket 52.0% → +3.6pp (ROI +6.9%) · GOOD

$ scripts/convert.sh 2.08 decimal hk,malay,american,probability
decimal=2.08 hk=1.08 malay=-0.926 american=108 probability=0.4808

The agent picks a verb and passes arguments; each verb prints one concise,
ready-to-read line. No account, no API key, no payment.

Why this exists

• Asian market data, normalized. Asian handicap, totals, and 1x2 (full-time
  and half-time) from 7 curated books, every quote convertible between decimal,
  Hong Kong, Malay, Indonesian, American, and implied probability — quarter-line
  math handled correctly.
• Built for small context windows. Every verb returns one line per item,
  capped (a busy weekend never floods the context). --detailed opts into full
  JSON only when you actually need it.
• Deterministic by design. The scripts own the endpoints; the agent never
  hand-builds an API call, guesses a URL, or web-searches a score. Ambiguity
  comes back as an explicit ask: with candidates — never a silent guess.
• Read-only, no betting advice. It reports lines, prices, and detected edges
  as data. It never places, recommends, or sizes a bet — and instructs the agent
  not to either.

Install

Claude Code — plugin (recommended)

/plugin marketplace add infersports/infersports-skill
/plugin install infersports@infersports

Prefer MCP? One line

claude mcp add --transport http infersports https://api.infersports.dev/mcp

The remote MCP server carries 16 tools (everything the skill does, plus per-book
depth, arbitrage detection, and opening lines). Skill and MCP server coexist fine.

Any agent (Codex · OpenClaw · others) — one prompt

Paste this whole block to your agent:

Install the "InferSports" sports-odds skill for me, verify it works, then show me what to ask.

1. Find where your agent loads Agent Skills and create it if missing:
   • Claude Code → ~/.claude/skills/   (or ./.claude/skills/ for this project only)
   • OpenClaw (lobster) → ~/.openclaw/workspace/skills/
   • Codex / other → your skills (agent-skills) directory
2. Clone this repo to a temp folder:
   git clone --depth 1 https://github.com/infersports/infersports-skill /tmp/infersports-skill
3. Find the folder containing SKILL.md (it's plugins/infersports/skills/infersports) and copy
   that whole folder — named `infersports` — into your skills directory.
4. chmod +x <skills_dir>/infersports/scripts/*.sh
5. Verify offline:  INFERSPORTS_MOCK=1 <skills_dir>/infersports/scripts/today.sh --limit 3   (→ 3 demo matches)
6. Verify live (KEYLESS — no key, no signup):  <skills_dir>/infersports/scripts/today.sh --limit 3   (→ real matches)
7. Reload skills if needed, then confirm install and give me 3 example questions.

Read-only & keyless — never needs an account, API key, or payment. Needs bash, curl, python3.

No git? Swap step 2 for: download https://infersports.dev/skill.tgz and extract
(tar xzf skill.tgz) so the infersports/ folder lands in your skills directory.

Requirements: bash, curl, python3 on PATH (already present on macOS/Linux).
The skill talks only to the public InferSports API at https://api.infersports.dev.

The 11 verbs

digest, preview, scan, and compare are packaged: each folds several
API reads (and the de-vig math) into a single capped answer, so the agent doesn't
burn context orchestrating the fan-out itself.

Match arguments accept a team name, "A vs B", or an evt_… id straight from
today/events output. Ambiguous names (senior vs U21, two same-day fixtures)
return an ask: list to pin down with --date / --sport.

What the data covers

Design notes

Things the skill does that a thin API wrapper wouldn't:

• Concise-by-default contract. One line per item, hard caps everywhere
  (today defaults to 20, max 50; digest ≤ 12). The skill is written for
  agents where every token of output competes with the user's actual task.
• ask: / ERROR:+FIX: protocol. Ambiguity and failures come back as
  structured, actionable lines the agent can act on mechanically — re-run with
  the suggested flag, or put the choice to the user. not found is documented
  as a final answer, not a cue to go scrape the web.
• Offline mode for CI. INFERSPORTS_MOCK=1 answers every verb from bundled
  fixtures/ — deterministic, no network. That's how you test an agent that
  uses this skill without flaky live data.
• No-pick policy, enforced in the prompt. SKILL.md instructs the agent to
  relay numbers and detected edges as printed and to decline "which bet should I
  place?" — detection is information, not advice.

Repo layout

plugins/infersports/skills/infersports/
  SKILL.md          # the manifest the agent reads: verbs, rules, examples
  scripts/          # the 11 verbs + shared _common.sh / _fmt.py
  references/       # full-api.md — the complete REST + MCP surface
  fixtures/         # offline mock responses (INFERSPORTS_MOCK=1)

Beyond the skill

The skill wraps the highest-frequency reads. The full API adds per-book odds
depth, cross-book line comparison, arbitrage detection, opening-line snapshots,
and the bookmaker catalogue:

• REST: https://api.infersports.dev · OpenAPI: https://api.infersports.dev/docs
• MCP (16 tools): claude mcp add --transport http infersports https://api.infersports.dev/mcp
• In-repo reference: references/full-api.md

Configuration (all optional)

License

MIT — see LICENSE.