Sports Context Protocol — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Sports Context Protocol (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
The context, safety, and memory layer for sports agents. Golf first.
Before a sports agent acts, it checks SCP. Then SCP learns from what happened.
SCP — Sports Context Protocol — is an open context layer for AI agents operating in sports. Every sport venue has the same five things underneath: inventory, rules, actions, consequences, and memory. SCP is the standard way an agent reads those before it acts, and learns from the outcome after.
SCP Golf is Profile 001 — the first working profile. Golf is the cleanest wedge because an agent cannot safely book, price, move, or recommend anything at a course without understanding tee-sheet state, protected inventory, pricing policy, pace risk, events, and operator memory. Golf makes the problem impossible to ignore.
This repository is SCP Golf Alpha: a synthetic demo course, a local MCP server, booking and pricing safety checks, soft holds, a decision ledger, and a self-learning memory. No real course data, no integrations, no database.
docs/SCP_CORE_SPEC.mddocs/SCP_PROFILES.mddocs/SCP_GOLF_PROFILE.mdAI golf agents are coming — answering calls, booking tee times, quoting prices, moving reservations. The problem: most agents only know the conversation. They do not know the course: the tee-sheet state, the member protections, the league blocks, the pricing floor, the pace risk, the operator's preferences, and what happened the last time a similar decision was made.
SCP Golf gives them that, and then it learns.
June 6, 2026: a 67-slot tee sheet with member, league, and outing blocks.
logging, outcome feedback, and learning insights.
overrides, pace issues, price rejections — so the next similar decision is better.
npm install
npm run build
npm run typecheck
npm run testnpm run dev # runs the MCP server on stdio (tsx, no build needed)
npm start # runs the compiled server from dist/Test it interactively with the MCP Inspector:
npx @modelcontextprotocol/inspector npm run dev| Tool | What it does |
|---|---|
get_course_context | Full operating context — read this before acting. |
get_available_inventory | Available tee times near a preferred time. |
check_booking_action | Is a booking allowed, blocked, risky? Writes a decision. |
check_pricing_action | Is a quoted/discounted price allowed? Writes a decision. |
create_soft_hold | Temporary hold on a tee time before confirmation. |
write_decision_event | Log a decision directly. |
submit_outcome_feedback | The learning tool. Feed an outcome back to SCP. |
get_learning_insights | What SCP has learned. |
explain_action | Explain a result for golfer / operator / developer. |
scp://course/demo and its children: context, tee-sheet, booking-policy, pricing-policy, events, weather, pace, decision-ledger, learning-memory, soft-holds.
This is the heart of SCP. It is operational learning — no model training.
deterministic description of the kind of decision).
submit_outcome_feedback.The demo moment: ask for Saturday ~09:00, have an operator override the result once, ask again — SCP now recommends the operator's preferred time. See docs/LEARNING_LOOP.md.
docs/SCP_CORE_SPEC.md — the protocol, sport-agnostic.docs/SCP_PROFILES.md — the profile system and roadmap.docs/SCP_GOLF_PROFILE.md — Profile 001 primitive mapping.docs/SCP_GOLF_SPEC.md — golf implementation detail.docs/QUICKSTART.md — run and test locally.docs/DEMO_PROMPTS.md — 10 demo prompts.docs/LEARNING_LOOP.md — how the learning works.docs/ROADMAP.md — phases beyond the alpha.Alpha. Synthetic data. Booking safety first. Self-learning from decision outcomes. Not partnered with any course, not integrated with any provider, not live with any operator.
MIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.