sai-perps

name	sai-perps
description	Query and reconcile Sai perps data across live Perp, Oracle, and SLP Vault contracts; Sai Keeper GraphQL; DexPal REST; sai-keeper Postgres; and referral records. Use for markets, open interest, fees, trades, vault health, oracle prices, REST statistics, database-backed metrics, referral ownership or history, trading-credit balances, app/indexer behavior, or mismatches between chain, APIs, databases, sheets, and operator records. Keep mutations and private operator workflows in sai-ops.
metadata	{"tags":["sai-perps","sai-keeper","sai-app"],"agent_skills":["nibiru-cli-nibid","postgresql-psql","sai-ops"]}

Sai Perps

Use this skill for Sai read/query/reconcile work across protocol state, indexed data, public APIs, database semantics, referral analytics, and app behavior.

Boundaries

Use agent skill sai-ops when a request uploads code, migrates contracts, executes admin messages, creates or updates referral codes, issues trading credits, changes production configuration, triggers trades, mutates a sheet, or writes a state-change changelog. This skill can supply read-only preflight and postflight checks.

For a Sai Keeper Postgres schema, SQL semantics, or freshness question, read references/sai-db.md. For direct private database access, Tailscale/primary-replica routing, or any write operation, also use the private agent skill sai-db; do not infer connection details from repository configuration or environment variables.

Reference router

Surface	Use when	Reference
Live contracts	Protocol state, contract configuration, markets, collaterals, OI limits, fees, vault mappings, deposits, referral ownership, and credit balances.	sai-contracts.md
SLP Vault contracts	Vault health, collateralization, share price, deposit caps, daily risk state, epochs, withdrawal timing, raw vault state, pending gov fees.	sai-vaults.md
Sai Keeper GraphQL	Indexed trades and LP history, app-visible oracle prices, fees, referral history, subscriptions, and user-facing entities.	sai-graphql.md
DexPal REST	Public product aggregates, stats, markets, yield, health checks, and date-window metrics.	sai-rest.md
Sai Keeper Postgres	Table and column meanings, units, migrations, routines, API implementation, and aggregate freshness.	sai-db.md
Cross-source reconciliation	Explaining why chain, GraphQL, REST, DB, sheets, or boku notes disagree.	sai-reconciliation.md
Referral checks	Existing referral code ownership, redemption history, sheet metadata, EVM/Bech32 normalization, testnet balance/credit checks.	sai-referrals.md

Diagnostic Order

For questions spanning multiple systems, prefer this order unless the prompt explicitly selects a surface:

1. Identify the network, entity, and source-of-truth layer: contract address, market index, token index, trader, vault, referral code, REST endpoint, GraphQL field, or database table.
2. Read live contract state when the fact controls protocol behavior.
3. Read GraphQL or REST when the fact is about what the app/API exposes.
4. Use database references or SQL when GraphQL/REST semantics, freshness, or aggregation logic are the question.
5. Compare sources explicitly. Say which layer owns each fact and which layer is stale, derived, human-maintained, or not expected to match exactly.

Common Pitfalls

• Contract query JSON uses CosmWasm snake_case; GraphQL fields use camelCase.
• On-chain and database amounts are often base units. GraphQL Int amounts are also base units. REST responses often expose human or USD values.
• TokenIndex(0) is the USD quote placeholder, not collateral. Mainnet collateral indices are TokenIndex(1) for USDC and TokenIndex(2) for stNIBI.
• Some contract queries require wrapped string indices like "MarketIndex(0)"; others require plain integers. Read the contract reference before guessing.
• REST 24-hour windows are not all identical. GET /dexpal/v1/stats and GET /dexpal/v1/markets/details can use different time anchors.
• Sheets and boku epics are operator records, not protocol truth.

Examples and trigger tuning

Use examples/sai-perps-prompts.md for realistic prompts that should trigger this skill, plus near misses that should use agent skill sai-ops or another specialized skill.