wisp-code-discipline

name	wisp-code-discipline
description	Hard constraints for work on wisp / sdk / connectors. Prefer existing types and call paths; ban parallel abstractions and recreated logic. Use whenever editing this monorepo, designing credentials/settings, packaging, lifecycle, or when the user says "don't reinvent", "existing API", "stay within boundaries", or "/wisp-code-discipline".

Wisp code discipline

Non-negotiable

1. Use what already exists. Before adding a type, interface, or package, grep for the same job (GetRequiredCredentialFields, NewConfig, Validate, MapToSDKConfig, StartStandalone, Wait, ResolveSettingsPath, etc.).
2. Banned: parallel abstractions. Do not invent CredentialSchemaProvider, second config loaders, second packaging paths, or “better” copies of SDK surfaces the CLI already calls.
3. Only add new code when absolutely needed — bugfix, test, or thin glue. Prefer delete/demote over extend.
4. Extend existing functions if discovery/validation is weak; do not replace them with a new public API of the same purpose.
5. Connectors own exchange config structs + Validate(). CLI/SDK discover fields via NewConfig() + existing GetRequiredCredentialFields / map+validate — not a new schema DSL.
6. No drive-by docs unless the user asked. No new markdown files for design notes.
7. Mocks via mockery, not hand stubs.
8. No DEPI (or other client/project) material in this codebase. Delete on sight. Never copy, generate, or leave DEPI* / depi.co.id / Sanctum-unrelated sales specs under wisp/sdk/connectors.

Credentials / Settings (current law)

Concern	Existing path
Global keys file	~/.wisp/connectors.yml via config.ResolveSettingsPath / Settings service
Field list for forms	ConnectorService.GetRequiredCredentialFields(exchange)
Config object	registry.Connector.NewConfig() then JSON map from credentials
Validate	MapToSDKConfig + Config.Validate()
Available venues	connectors/types.AllConnectors

Do not add CredentialField / CredentialSchemaProvider unless the user explicitly orders a redesign after rejecting the existing path.

Packaging (current law)

• Standalone only: main.go + StartStandalone + Wait
• Plugins / run-strategy / .so are removed — do not reintroduce

Trading signals (current law)

Path	Purpose
wisp.Spot().Emit / Perp().Emit / Predict().Emit / Options().Emit	Only way to place orders (market-scoped)
BaseStrategy.EmitStatus	Operator/monitor status snapshots — not trading

No wisp.Emit. No signal publish channel.

Config boot flow (current law)

strategyDir/config.yml     → StrategyConfig (exchanges/assets, no secrets)
ResolveSettingsPath        → ~/.wisp/connectors.yml (Configuration)
GetConnectorConfigsForStrategy → MapToSDKConfig + Validate → connector.Config
StartStandalone            → Initialize connectors + lifecycle

Redundancy sweep

cd sdk && make redundancy   # or: go run ./tools/redundancy

Uses product-surface root (wisp blank-import) + structural checks. Prefer deleting zombies before adding markets. Mocks: mockery only.

Clarity / redundancy backlog

Market shell (standard layout): markets/<domain>/{module,facade,types,signal,executor,store,ingestor} — clone that tree; domain Module Provides facade.New*; add Module to pkg/modules.go.

Done: instruments YAML removed; hard-fail unregistered connectors at init; spot/perp pair shell shared via markets/base/{watchlist,analytics,universe,views,activity,facade,executor,store,signal,types}; PairAction + PairSignalData + builder Core; domain only wraps fluent types + leverage/funding; package pkg (was packages); dead spot/perp config/ packages removed.

Still open: options/prediction shells (different keys); optional pkg/testing trim.

Released: sdk v0.1.10 (pair shell DRY + config honesty).

Process: no release tags unless ship-worthy change.

When tempted to invent

1. Name the existing type/function that already does it.
2. If incomplete, improve that implementation + tests.
3. If truly missing, state the gap in one sentence and get confirmation before a new public type.

Tests

Add tests against existing packages (compile, settings, manager, connector service). Do not invent test-only APIs that production never uses.