agentsclimarketplace

Go error handling

Skill cxuu/golang-skills/skills/go-error-handling

Use when writing Go code that returns, wraps, or handles errors — choosing between sentinel errors, custom types, and fmt.Errorf (%w vs %v), structuring error flow, or deciding whether to log or return. Also use when propagating errors across package boundaries or using errors.Is/As, even if the user doesn't ask about error strategy. Does not cover panic/recover patterns (see go-defensive).From its SKILL.md

Install
npx -y skills add cxuu/golang-skills --skill go-error-handling

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • runs commandsInstructs the agent to run 2 commands, including `bash scripts/check-errors.sh` and 1 more.

SKILL.md

5.9 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it

Go Error Handling

Compatibility: errors.Is, errors.As, and %w wrapping require Go 1.13+; structured logging examples may use log/slog from Go 1.21+.

Resource Routing

  • scripts/check-errors.sh - Run when checking string-based error matching, bare error propagation, and log-and-return patterns.
  • scripts/check-errors-ast.go - Implementation helper invoked by check-errors.sh; patch this when changing error-flow analysis behavior.
  • references/ERROR-FLOW.md - Read when deciding where to handle, wrap, log, or return errors.
  • references/ERROR-TYPES.md - Read when choosing sentinel errors, typed errors, or opaque errors.
  • references/WRAPPING.md - Read when choosing %w versus %v or crossing package boundaries.

In Go, errors are values — they are created by code and consumed by code.

Choosing an Error Strategy

  1. System boundary (RPC, IPC, storage)? → Wrap with %v to avoid leaking internals
  2. Caller needs to match specific conditions? → Sentinel or typed error, wrap with %w
  3. Caller just needs debugging context? → fmt.Errorf("...: %w", err)
  4. Leaf function, no wrapping needed? → Return the error directly

Default: wrap with %w and place it at the end of the format string.


Core Rules

Never Return Concrete Error Types

Never return concrete error types from exported functions — a concrete nil pointer can become a non-nil interface:

// Bad: Concrete type can cause subtle bugs
func Bad() *os.PathError { /*...*/ }

// Good: Always return the error interface
func Good() error { /*...*/ }

Error Strings

Error strings should not be capitalized and should not end with punctuation. Exception: exported names, proper nouns, or acronyms.

// Bad
err := fmt.Errorf("Something bad happened.")

// Good
err := fmt.Errorf("something bad happened")

For displayed messages (logs, test failures, API responses), capitalization is appropriate.

Return Values on Error

When a function returns an error, callers must treat all non-error return values as unspecified unless explicitly documented.

Tip: Functions taking a context.Context should usually return an error so callers can determine if the context was cancelled.


Handling Errors

When encountering an error, make a deliberate choice — do not discard with _:

  1. Handle immediately — address the error and continue
  2. Return to caller — optionally wrapped with context
  3. In exceptional caseslog.Fatal or panic

To intentionally ignore: add a comment explaining why.

n, _ := b.Write(p) // never returns a non-nil error

For related concurrent operations, use errgroup:

g, ctx := errgroup.WithContext(ctx)
g.Go(func() error { return task1(ctx) })
g.Go(func() error { return task2(ctx) })
if err := g.Wait(); err != nil { return err }

Avoid In-Band Errors

Don't return -1, nil, or empty string to signal errors. Use multiple returns:

// Bad: In-band error value
func Lookup(key string) int  // returns -1 for missing

// Good: Explicit error or ok value
func Lookup(key string) (string, bool)

This prevents callers from writing Parse(Lookup(key)) — it causes a compile-time error since Lookup(key) has 2 outputs.


Error Flow

Handle errors before normal code. Early returns keep the happy path unindented:

// Good: Error first, normal code unindented
if err != nil {
    return err
}
// normal code

Handle errors once — either log or return, never both:

Error encountered?
├─ Caller can act on it? → Return (with context via %w)
├─ Top of call chain? → Log and handle
└─ Neither? → Log at appropriate level, continue

Error Types

Advisory: Recommended best practice.

Caller needs to match?Message typeUse
Nostaticerrors.New("message")
Nodynamicfmt.Errorf("msg: %v", val)
Yesstaticvar ErrFoo = errors.New("...")
Yesdynamiccustom error type

Default: Wrap with fmt.Errorf("...: %w", err). Escalate to sentinels for errors.Is(), to custom types for errors.As().


Error Wrapping

Advisory: Recommended best practice.

  • Use %v: At system boundaries, for logging, to hide internal details
  • Use %w: To preserve error chain for errors.Is/errors.As

Key rules: Place %w at the end. Add context callers don't have. If annotation adds nothing, return err directly.

Validation: After implementing error handling, run bash scripts/check-errors.sh to detect common anti-patterns. Then run go vet ./... to catch additional issues.


Related Skills

  • Error naming: See go-naming when naming sentinel errors (ErrFoo) or custom error types
  • Testing errors: See go-testing when testing error semantics with errors.Is/errors.As or writing error-checking helpers
  • Panic handling: See go-defensive when deciding between panic and error returns, or writing recover guards
  • Guard clauses: See go-control-flow when structuring early-return error flow or reducing nesting
  • Logging decisions: See go-logging when choosing log levels, configuring structured logging, or deciding what context to include in log messages

What ships with it: 5 files

23.2 KB alongside SKILL.md, 1 of them executable

references/

scripts/

Gives 0 of the 12 instructions most error diagnosis skills give in ~1.4k tokens

Counted across 135 of the 162 authors here whose files we hold, read 2026-09-06

  • Handle, re-throw, or log in every catch blockin 12 of 135, across 7 files
  • Use typed error classes over string messagesin 11 of 135, across 6 files
  • Log full error context server-sidein 10 of 135, across 5 files
  • Document every error code clients may receivein 9 of 135, across 4 files
  • Surface errors at the boundary where they occurin 9 of 135, across 4 files
  • Wrap React components in an ErrorBoundaryin 9 of 135, across 4 files
  • Wrap errors with context, never lose the originalin 9 of 135, across 4 files
  • Use the standard error envelope for API responsesin 9 of 135, across 4 files
  • Retry only retriable errors, never 4xx client errorsin 8 of 135, across 3 files
  • Retry transient failures with exponential backoff and jitterin 8 of 135
  • Show users friendly messages without technical detailsin 7 of 135, across 3 files
  • Use the Result pattern for expected failuresin 7 of 135, across 5 files

Said here and by no other author read

  • Use %w to preserve the error chain for errors.Is/As
  • Add context callers don't have when wrapping
  • Return err directly when wrapping adds nothing
  • Handle, return, or deliberately ignore every error
  • Comment when intentionally ignoring an error
  • Handle errors before normal code with early returns

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.