| name | Haskell Ecosystem |
| description | This skill should be used when working with Haskell projects, "cabal.project", "stack.yaml", "ghc", "cabal build/test/run", "stack build/test/run", or Haskell language patterns. Provides comprehensive Haskell ecosystem patterns and best practices. |
| version | 2.0.0 |
Provide comprehensive patterns for Haskell language, GHC toolchain, Cabal/Stack build systems, and type-level programming.
Read - Analyze cabal files, stack.yaml, and Haskell source files
Edit - Modify Haskell code and build configuration
Bash - Run cabal build, stack build, ghci commands
mcp__context7__get-library-docs - Fetch latest Haskell documentation
Functions have no side effects; same input always produces same output
Expressions evaluated only when needed; enables infinite data structures
Hindley-Milner type system; explicit signatures recommended for top-level definitions
Compose computations with effects; IO, Maybe, Either, State, Reader, Writer
Ad-hoc polymorphism; define operations for types (Eq, Ord, Show, Functor, Applicative, Monad)
<haskell_language>
<type_system>
Sum types (Either, Maybe) and product types (tuples, records)
data Maybe a = Nothing | Just a
data Either a b = Left a | Right b
data Person = Person { name :: Text, age :: Int }
<concept name="type_classes">
<description>Define behavior for types; similar to interfaces but more powerful</description>
<common_classes>
<class name="Eq">Equality comparison (==, /=)</class>
<class name="Ord">Ordering comparison (compare, <, >, <=, >=)</class>
<class name="Show">String representation (show)</class>
<class name="Read">Parse from string (read)</class>
<class name="Functor">Mappable containers (fmap, <$>)</class>
<class name="Applicative">Apply functions in context (<*>, pure)</class>
<class name="Monad">Sequence computations (>>=, return, do-notation)</class>
<class name="Foldable">Reducible structures (foldr, foldl, toList)</class>
<class name="Traversable">Traverse with effects (traverse, sequenceA)</class>
<class name="Semigroup">Associative binary operation (<>)</class>
<class name="Monoid">Semigroup with identity (mempty, mappend)</class>
</common_classes>
<example>
class Eq a where
(==) :: a -> a -> Bool
(/=) :: a -> a -> Bool
instance Eq Person where
p1 == p2 = p1.name == p2.name && p1.age == p2.age
</example>
</concept>
<concept name="type_level_programming">
<description>Advanced type system features for compile-time guarantees</description>
<pattern name="gadts">
<description>Generalized Algebraic Data Types - refine types in pattern matching</description>
<pragma>GADTs</pragma>
<example>
{-# LANGUAGE GADTs #-}
data Expr a where
LitInt :: Int -> Expr Int
LitBool :: Bool -> Expr Bool
Add :: Expr Int -> Expr Int -> Expr Int
If :: Expr Bool -> Expr a -> Expr a -> Expr a
eval :: Expr a -> a
eval (LitInt n) = n
eval (LitBool b) = b
eval (Add e1 e2) = eval e1 + eval e2
eval (If c t e) = if eval c then eval t else eval e
</example>
</pattern>
<pattern name="type_families">
<description>Type-level functions; compute types from types</description>
<pragma>TypeFamilies</pragma>
<example>
{-# LANGUAGE TypeFamilies #-}
type family Element c where
Element [a] = a
Element (Set a) = a
Element Text = Char
class Container c where
type Elem c
empty :: c
insert :: Elem c -> c -> c
</example>
</pattern>
<pattern name="data_kinds">
<description>Promote data types to kinds for type-level programming</description>
<pragma>DataKinds</pragma>
<example>
{-# LANGUAGE DataKinds, KindSignatures #-}
data Nat = Zero | Succ Nat
data Vec (n :: Nat) a where
VNil :: Vec 'Zero a
VCons :: a -> Vec n a -> Vec ('Succ n) a
</example>
</pattern>
<pattern name="phantom_types">
<description>Type parameters that don't appear in value constructors</description>
<example>
newtype Tagged tag a = Tagged { unTagged :: a }
data Validated
data Unvalidated
validateEmail :: Tagged Unvalidated Text -> Maybe (Tagged Validated Text)
</example>
</pattern>
</concept>
<concept name="higher_kinded_types">
<description>Type constructors as parameters; enables Functor, Monad abstractions</description>
<example>
class Functor f where
fmap :: (a -> b) -> f a -> f b
-- f is a type constructor (* -> *)
-- Functor operates on type constructors, not concrete types
</example>
</concept>
</type_system>
<monad_transformers>
Compose monadic effects using transformers from mtl/transformers
<pattern name="transformers_stack">
<description>Build custom monad stacks with transformers</description>
<example>
import Control.Monad.Trans.Reader
import Control.Monad.Trans.State
import Control.Monad.Trans.Except
type App a = ReaderT Config (StateT AppState (ExceptT AppError IO)) a
runApp :: Config -> AppState -> App a -> IO (Either AppError (a, AppState))
runApp cfg st app = runExceptT (runStateT (runReaderT app cfg) st)
</example>
</pattern>
<pattern name="mtl_style">
<description>Use mtl type classes for polymorphic effect handling</description>
<example>
import Control.Monad.Reader
import Control.Monad.State
import Control.Monad.Except
doSomething :: (MonadReader Config m, MonadState AppState m, MonadError AppError m) => m Result
doSomething = do
cfg <- ask
st <- get
when (invalid cfg) $ throwError InvalidConfig
pure (compute cfg st)
</example>
</pattern>
<common_transformers>
<transformer name="ReaderT">Read-only environment; ask, local</transformer>
<transformer name="StateT">Mutable state; get, put, modify</transformer>
<transformer name="ExceptT">Error handling; throwError, catchError</transformer>
<transformer name="WriterT">Accumulate output; tell, listen</transformer>
<transformer name="MaybeT">Short-circuit on Nothing</transformer>
<transformer name="ListT">Non-determinism (use list-t or logict for correct semantics)</transformer>
</common_transformers>
</monad_transformers>
<effect_systems>
Modern alternatives to mtl monad transformer stacks
<system name="effectful">
<description>Fast extensible effects library; recommended modern alternative to mtl for new projects</description>
<example>
import Effectful
import Effectful.Reader.Static
import Effectful.State.Static.Local
import Effectful.Error.Static
doSomething :: (Reader Config :> es, State AppState :> es, Error AppError :> es) => Eff es Result
doSomething = do
cfg <- ask
st <- get
when (invalid cfg) $ throwError InvalidConfig
pure (compute cfg st)
runApp :: Config -> AppState -> Eff '[Reader Config, State AppState, Error AppError, IOE] a -> IO (Either (CallStack, AppError) a)
runApp cfg st = runEff . runError . runState st . runReader cfg
</example>
<note>Better performance than mtl due to no newtype wrapping; better error messages; supports static and dynamic dispatch</note>
</system>
<system name="bluefin">
<description>Effect system using linearity for safety; alternative to effectful</description>
<note>Uses a different approach based on handles rather than type-level effect lists</note>
</system>
<decision_tree name="effect_system_choice">
<question>Which effect system should I use?</question>
<branch condition="Existing mtl codebase">Keep mtl; migrate incrementally if needed</branch>
<branch condition="New project, performance matters">effectful (best performance)</branch>
<branch condition="Simple application">mtl is still fine</branch>
<branch condition="Experimental, novel approach">bluefin</branch>
</decision_tree>
</effect_systems>
<linear_types>
LinearTypes extension (stable since GHC 9.0); ensures values are used exactly once
{-# LANGUAGE LinearTypes #-}
-- A linear function: argument must be used exactly once
dup :: a %1 -> (a, a) -- This would NOT type-check
-- Useful for resource management
withFile :: FilePath -> (Handle %1 -> IO a) %1 -> IO a
</example>
<note>Primarily useful for resource-safe APIs and performance-critical code. Not needed for most application code.</note>
</linear_types>
Composable getters, setters, and traversals using lens library
<pattern name="lens_basics">
<description>Focus on parts of data structures</description>
<example>
import Control.Lens
data Person = Person { _name :: Text, _age :: Int }
makeLenses ''Person
-- name :: Lens' Person Text
-- age :: Lens' Person Int
getName :: Person -> Text
getName p = p ^. name
setName :: Text -> Person -> Person
setName n p = p & name .~ n
modifyAge :: (Int -> Int) -> Person -> Person
modifyAge f p = p & age %~ f
</example>
</pattern>
<pattern name="optic_types">
<description>Different optic types for different access patterns</description>
<optics_hierarchy>
<optic name="Lens">Get and set exactly one value</optic>
<optic name="Prism">Focus on one branch of a sum type</optic>
<optic name="Traversal">Focus on zero or more values</optic>
<optic name="Iso">Bidirectional transformation</optic>
<optic name="Getter">Read-only access</optic>
<optic name="Fold">Read-only traversal</optic>
</optics_hierarchy>
</pattern>
<pattern name="lens_operators">
<description>Common lens operators</description>
<operators>
<operator name="^.">View through lens (view)</operator>
<operator name=".~">Set value (set)</operator>
<operator name="%~">Modify value (over)</operator>
<operator name="&">Apply function (flip ($))</operator>
<operator name="^?">View through prism (preview)</operator>
<operator name="^..">View all through traversal (toListOf)</operator>
</operators>
</pattern>
<alternative name="optics">
<description>Alternative to lens with better type errors and composition</description>
<package>optics</package>
<note>Considered more modern; lens has larger ecosystem</note>
</alternative>
<error_handling>
Optional values; prefer over null
findUser :: UserId -> Maybe User
findUser uid = lookup uid users
-- Safe chaining with Monad
getUserEmail :: UserId -> Maybe Email
getUserEmail uid = do
user <- findUser uid
pure (userEmail user)
</example>
</pattern>
<pattern name="either">
<description>Computations that may fail with error information</description>
<example>
parseConfig :: Text -> Either ParseError Config
parseConfig input = do
json <- parseJSON input
validateConfig json
</example>
</pattern>
<pattern name="exceptt">
<description>Error handling in monadic contexts</description>
<example>
import Control.Monad.Except
data AppError = NotFound | InvalidInput Text | IOError IOException
loadUser :: MonadError AppError m => UserId -> m User
loadUser uid = do
mUser <- findUser uid
case mUser of
Nothing -> throwError NotFound
Just u -> pure u
</example>
</pattern>
</error_handling>
<common_patterns>
Zero-cost wrapper for type safety; use DerivingStrategies and DerivingVia
{-# LANGUAGE DerivingStrategies #-}
{-# LANGUAGE DerivingVia #-}
newtype UserId = UserId Int
deriving stock (Show)
deriving newtype (Eq, Ord)
newtype Email = Email Text
deriving stock (Show)
deriving newtype (Eq, IsString)
-- DerivingVia for custom instances
newtype Dollars = Dollars Int
deriving (Semigroup, Monoid) via (Sum Int)
</example>
</pattern>
<pattern name="smart_constructors">
<description>Validate data at construction time</description>
<example>
module Email (Email, mkEmail, unEmail) where
newtype Email = Email Text
mkEmail :: Text -> Maybe Email
mkEmail t
| isValidEmail t = Just (Email t)
| otherwise = Nothing
</example>
</pattern>
<pattern name="records">
<description>Modern record pattern using NoFieldSelectors + OverloadedRecordDot (since GHC 9.2)</description>
<example>
{-# LANGUAGE NoFieldSelectors #-}
{-# LANGUAGE OverloadedRecordDot #-}
data Config = Config
{ host :: Text
, port :: Int
, timeout :: Int
}
-- Using OverloadedRecordDot (dot syntax for field access)
mkConnection :: Config -> IO Connection
mkConnection cfg = connect cfg.host cfg.port
-- Record update syntax still works
withTimeout :: Int -> Config -> Config
withTimeout t cfg = cfg { timeout = t }
</example>
<note>NoFieldSelectors prevents generating top-level field accessor functions, avoiding name clashes. OverloadedRecordDot enables dot-syntax field access. This is the modern standard replacing prefixed field names (configHost, configPort).</note>
</pattern>
</common_patterns>
<anti_patterns>
Functions that crash on some inputs (head, tail, fromJust, read)
Use safe alternatives (headMay, listToMaybe) or pattern matching
<avoid name="string_type">
<description>Using String ([Char]) for text processing</description>
<instead>Use Text or ByteString for performance</instead>
</avoid>
<avoid name="lazy_io">
<description>Using lazy IO (readFile, getContents) in production</description>
<instead>Use strict IO or streaming (conduit, pipes, streaming)</instead>
</avoid>
<avoid name="orphan_instances">
<description>Defining type class instances outside the module of the type or class</description>
<instead>Use newtype wrappers or define instances in appropriate modules</instead>
</avoid>
<avoid name="existential_overuse">
<description>Overusing ExistentialQuantification to simulate OOP-style polymorphism</description>
<instead>Use sum types, type classes with concrete types, or the Handle pattern. Existentials hide type information and make code harder to reason about.</instead>
<example_bad>
-- Anti-pattern: existential wrapper for heterogeneous collection
data AnyShow = forall a. Show a => AnyShow a
items :: [AnyShow]
items = [AnyShow (1 :: Int), AnyShow "hello"]
</example_bad>
<example_good>
-- Prefer: explicit sum type
data Item = ItemInt Int | ItemText Text
deriving stock (Show)
</example_good>
</avoid>
<avoid name="old_style_records">
<description>Using prefixed field names (configHost, personName) to avoid name clashes</description>
<instead>Use NoFieldSelectors + OverloadedRecordDot for modern record access</instead>
</avoid>
</anti_patterns>
</haskell_language>
.
├── my-project.cabal
├── cabal.project # Multi-package configuration
├── cabal.project.local # Local overrides (gitignored)
├── src/
│ └── MyProject.hs
├── app/
│ └── Main.hs
├── test/
│ └── Spec.hs
└── bench/
└── Bench.hs
<cabal_file>
<basic_structure>
cabal-version: 3.0
name: my-project
version: 0.1.0.0
synopsis: Short description
license: MIT
author: Your Name
maintainer: your@email.com
common warnings
ghc-options: -Wall -Wcompat -Widentities -Wincomplete-record-updates
-Wincomplete-uni-patterns -Wpartial-fields -Wredundant-constraints
library
import: warnings
exposed-modules: MyProject
build-depends: base ^>=4.21,
text ^>=2.0,
containers ^>=0.6
hs-source-dirs: src
default-language: GHC2024
executable my-project
import: warnings
main-is: Main.hs
build-depends: base ^>=4.21,
my-project
hs-source-dirs: app
default-language: GHC2024
test-suite my-project-test
import: warnings
type: exitcode-stdio-1.0
main-is: Spec.hs
build-depends: base ^>=4.21,
my-project,
hspec ^>=2.11,
QuickCheck ^>=2.14
hs-source-dirs: test
default-language: GHC2024
</basic_structure>
<version_bounds>
-- Caret (^>=): major version compatible
base ^>=4.21 -- 4.21.x.x (GHC 9.14)
-- Range
text >=2.0 && <2.2
-- Any version (avoid in published packages)
containers
</version_bounds>
<language_extensions>
default-extensions:
OverloadedStrings
OverloadedRecordDot
NoFieldSelectors
LambdaCase
DerivingStrategies
DerivingVia
TypeFamilies
default-language: GHC2024 -- Recommended for new projects (default since GHC 9.12)
</language_extensions>
</cabal_file>
<cabal_project>
Multi-package project configuration
packages: .
./subpackage
-- Use local package
optional-packages: ../local-dependency
-- Optimization
optimization: 2
-- Documentation
documentation: True
-- Test options
tests: True
-- Allow newer dependencies
allow-newer: base
</example>
</cabal_project>
Compile the project
Build all targets
Build and run executable
Run test suites
Start GHCi with project loaded
Generate documentation
Update package index
Lock dependency versions
Generate version bounds
Check for outdated dependencies
<cabal_install>
<current_version>cabal-install 3.14+</current_version>
Use cabal.project files for multi-package projects and dependency configuration
</cabal_install>
.
├── stack.yaml # Stack project configuration
├── stack.yaml.lock # Locked dependency versions
├── package.yaml # hpack format (generates .cabal)
└── ... (same as cabal)
<stack_yaml>
<basic_structure>
resolver: lts-23.0 # Stackage LTS snapshot
packages:
- .
- ./subpackage
extra-deps:
- some-package-1.0.0
- github: owner/repo
commit: abc123
ghc-options:
"$locals": -Wall -Werror
</basic_structure>
</stack_yaml>
<package_yaml>
hpack format; generates .cabal file
name: my-project
version: 0.1.0.0
dependencies:
- base >= 4.21 && < 5
- text
- containers
ghc-options:
- -Wall
- -Wcompat
default-extensions:
- OverloadedStrings
- LambdaCase
library:
source-dirs: src
executables:
my-project:
main: Main.hs
source-dirs: app
dependencies:
- my-project
tests:
my-project-test:
main: Spec.hs
source-dirs: test
dependencies:
- my-project
- hspec
- QuickCheck
</example>
</package_yaml>
Compile the project
Run tests
Build and run executable
Start GHCi with project loaded
Generate documentation
Clean build artifacts
Upgrade Stack itself
<decision_tree name="cabal_vs_stack">
Which build tool should I use?
Stack with LTS resolver
Cabal (native format)
Cabal with cabal.project
Stack (simpler getting started)
Cabal with haskell.nix or nixpkgs
</decision_tree>
Glasgow Haskell Compiler
GHC 9.14.1 (latest major/LTS), GHC 9.12.4 (latest 9.12 bugfix)
<common_options>
<option name="-Wall">Enable all warnings</option>
<option name="-Werror">Treat warnings as errors</option>
<option name="-O2">Optimization level 2</option>
<option name="-threaded">Enable threaded runtime</option>
<option name="-rtsopts">Enable runtime system options</option>
<option name="-with-rtsopts">Set default RTS options</option>
</common_options>
<language_standards>
<standard name="Haskell2010">Legacy standard; avoid for new projects</standard>
<standard name="GHC2021">Widely supported; enables common extensions</standard>
<standard name="GHC2024" recommended="true">Current recommended default (since GHC 9.12); extends GHC2021 with ExplicitNamespaces, TypeData, MonoLocalBinds, and others</standard>
</language_standards>
<ghc_lts_policy>
<description>Starting with GHC 9.14, designated major versions receive LTS status with minimum 2 years of bugfix support. No new features are backported to LTS releases.</description>
<current_lts>GHC 9.14 (first LTS release)</current_lts>
</ghc_lts_policy>
<release_cadence>Major releases are made twice a year.</release_cadence>
Haskell Language Server - IDE support via LSP
Code completion
Type information on hover
Go to definition
Find references
Code actions (import, qualify)
Diagnostics (errors, warnings, hlint)
Code formatting (fourmolu, ormolu, stylish-haskell)
<configuration>
<file_reference>hls.yaml or hie.yaml</file_reference>
cradle:
cabal:
- path: "src"
component: "lib:my-project"
- path: "app"
component: "exe:my-project"
- path: "test"
component: "test:my-project-test"
</configuration>
Standard Haskell formatter; ormolu fork with more configuration options. Recommended for new projects.
fourmolu -i src/**/*.hs
fourmolu.yaml
<formatter name="ormolu">
<description>Minimal configuration formatter; fourmolu is preferred for most projects</description>
<usage>ormolu -i src/**/*.hs</usage>
</formatter>
<formatter name="stylish-haskell" status="legacy">
<description>Configurable import organization and formatting; largely superseded by fourmolu</description>
<usage>stylish-haskell -i src/**/*.hs</usage>
</formatter>
Suggest idiomatic Haskell improvements
hlint src/
.hlint.yaml
- ignore: {name: "Use camelCase"}
- warn: {name: "Use head"}
- error: {name: "Use String"}
<linter name="stan">
<description>Static analysis for Haskell</description>
<usage>stan</usage>
</linter>
<linter name="weeder">
<description>Detect dead code</description>
<usage>weeder</usage>
</linter>
BDD-style testing framework
import Test.Hspec
main :: IO ()
main = hspec $ do
describe "Calculator" $ do
it "adds two numbers" $ do
add 1 2 `shouldBe` 3
it "handles negative numbers" $ do
add (-1) 1 `shouldBe` 0
describe "Parser" $ do
context "when input is valid" $ do
it "parses successfully" $ do
parse "valid" `shouldSatisfy` isRight
context "when input is invalid" $ do
it "returns error" $ do
parse "invalid" `shouldSatisfy` isLeft
</example>
<matchers>
<matcher name="shouldBe">Equality check</matcher>
<matcher name="shouldSatisfy">Predicate check</matcher>
<matcher name="shouldReturn">IO action result</matcher>
<matcher name="shouldThrow">Exception check</matcher>
<matcher name="shouldContain">List containment</matcher>
</matchers>
Property-based testing; generate random test cases
import Test.QuickCheck
prop_reverseReverse :: [Int] -> Bool
prop_reverseReverse xs = reverse (reverse xs) == xs
prop_sortIdempotent :: [Int] -> Bool
prop_sortIdempotent xs = sort (sort xs) == sort xs
-- With preconditions
prop_headLast :: NonEmptyList Int -> Bool
prop_headLast (NonEmpty xs) = head xs == head xs
-- Custom generators
newtype PositiveInt = PositiveInt Int deriving Show
instance Arbitrary PositiveInt where
arbitrary = PositiveInt . abs <$> arbitrary
</example>
<integration>
-- With HSpec
describe "reverse" $ do
it "is its own inverse" $ property $
\xs -> reverse (reverse xs) == (xs :: [Int])
</integration>
Flexible test framework that unifies HUnit, QuickCheck, and other test providers under one runner
import Test.Tasty
import Test.Tasty.HUnit
import Test.Tasty.QuickCheck as QC
main :: IO ()
main = defaultMain tests
tests :: TestTree
tests = testGroup "Tests"
[ testCase "Addition" $ 1 + 1 @?= 2
, QC.testProperty "reverse/reverse" $ \xs ->
reverse (reverse xs) == (xs :: [Int])
]
</example>
<note>Good choice when combining multiple test providers; hspec is better for BDD-style tests</note>
Modern property-based testing with integrated shrinking
import Hedgehog
import qualified Hedgehog.Gen as Gen
import qualified Hedgehog.Range as Range
prop_reverse :: Property
prop_reverse = property $ do
xs <- forAll $ Gen.list (Range.linear 0 100) Gen.alpha
reverse (reverse xs) === xs
</example>
<best_practices>
Use QuickCheck for properties; HSpec for examples
Test edge cases: empty lists, zero, negative numbers
Use type-driven development; write types first
Use doctest for documentation examples
</best_practices>
<context7_integration>
Use Context7 MCP for up-to-date Haskell documentation
<haskell_libraries>
</haskell_libraries>
<usage_patterns>
resolve-library-id libraryName="optics haskell"
Workflow guidance
Step completed
get-library-docs context7CompatibleLibraryID="/websites/hackage_haskell_package_optics-0_4_2_1" topic="lenses"
Workflow guidance
Step completed
<pattern name="testing_framework">
<step order="1">
get-library-docs context7CompatibleLibraryID="/websites/hackage_haskell_package_hspec-2_11_12" topic="expectations"
Workflow guidance
Step completed
<pattern name="web_framework">
<step order="1">
get-library-docs context7CompatibleLibraryID="/haskell-servant/servant" topic="server"
Workflow guidance
Step completed
</usage_patterns>
</context7_integration>
<best_practices>
Let types guide design; use the type system to prevent errors
Avoid partial functions; use safe alternatives
Run hlint and fix suggestions before committing
Use Text/ByteString instead of String for performance
Prefer mtl-style or effectful constraints over concrete monad stacks
Write property-based tests for core logic
Document exported functions with Haddock comments
Use newtypes for type safety
Enable GHC2024 as default-language for new projects
Use NoFieldSelectors + OverloadedRecordDot for records
Use DerivingStrategies and DerivingVia for deriving instances
</best_practices>
Run hlint before committing; address all suggestions
Avoid partial functions (head, tail, fromJust); use safe alternatives
Use types to encode invariants; make illegal states unrepresentable
Use fourmolu for consistent formatting (standard formatter)
Prefer Text over String for text processing
Write HSpec tests for behavior; QuickCheck for properties
Use explicit type signatures for top-level definitions
Understand Haskell code requirements
1. Check cabal file or package.yaml for project configuration
Workflow guidance
Step completed
2. Review existing types and type classes
Workflow guidance
Step completed
3. Identify monad transformer requirements
Workflow guidance
Step completed
4. Check for type-level programming needs
Workflow guidance
Step completed
Write pure, type-safe Haskell code
1. Design with types first; let types guide implementation
Workflow guidance
Step completed
2. Use appropriate abstractions (Functor, Applicative, Monad)
Workflow guidance
Step completed
3. Handle errors with Maybe/Either/ExceptT
Workflow guidance
Step completed
4. Write property-based tests for core logic
Workflow guidance
Step completed
Verify Haskell code correctness
1. Run cabal build or stack build
Workflow guidance
Step completed
2. Run hlint for suggestions
Workflow guidance
Step completed
3. Run cabal test or stack test
Workflow guidance
Step completed
4. Check formatting with fourmolu/ormolu
Workflow guidance
Step completed
<error_escalation>
HLint suggestion about style
Apply suggestion, maintain idiomatic code
Type error or missing instance
Review types, add instance or adjust design
Breaking change in public API
Stop, present migration options to user
Partial function usage or unsafe code in library
Block operation, require safe alternatives
</error_escalation>
Use types to encode invariants
Avoid partial functions in library code
Follow Haskell style conventions
Using String for text processing
Lazy IO in production code
Orphan instances
<related_skills>
Navigate type class hierarchies and module structure
Fetch Haskell documentation and library references
Debug type errors, missing instances, and performance issues
haskell.nix integration and nixpkgs Haskell infrastructure
</related_skills>