fail-early-pattern
À propos
Cette compétence apprend aux développeurs à mettre en œuvre le principe "fail-fast" en validant les entrées et en signalant les erreurs immédiatement grâce à des clauses de garde et des assertions. Elle fournit des exemples centrés sur R et des conseils polyglottes pour écrire des fonctions robustes, sécuriser des API et refactoriser du code sujet aux erreurs. Utilisez-la lors de la réception de données externes, de la préparation d'une soumission CRAN ou de la revue de code pour éviter les défaillances silencieuses.
Installation rapide
Claude Code
Recommandénpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/fail-early-patternCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Fail Early
If fails → fail early, loud, w/ context. Codifies: validate inputs at boundaries, guard clauses reject bad state before propagates, err msgs answer what failed, where, why, how to fix.
Use When
- Writing/reviewing fns accepting external input (user data, API, file)
- Input validation before CRAN submission
- Refactor silent wrong results → errors
- Review PRs for err-handling quality
- Harden internal APIs vs invalid args
In
- Required: Fn/module to apply pattern
- Required: Trust boundaries (where external data enters)
- Optional: Existing err-handling to refactor
- Optional: Target language (default R; also Python, TypeScript, Rust)
Do
Step 1: Trust Boundaries
Map external data entry. Points needing validation:
- Public API fns (exported package)
- User-facing params
- File I/O (configs, data, uploads)
- Net responses (APIs, DBs)
- Env vars + system config
Internal helpers called only by validated code generally no redundant validation.
→ List entry points where untrusted data crosses.
If err: unclear → trace backwards from err logs/bug reports → find where bad data entered.
Step 2: Guard Clauses at Entry
Validate at top of each public fn before work.
R (base):
calculate_summary <- function(data, method = c("mean", "median", "trim"), trim_pct = 0.1) {
# Guard: type check
if (!is.data.frame(data)) {
stop("'data' must be a data frame, not ", class(data)[[1]], call. = FALSE)
}
# Guard: non-empty
if (nrow(data) == 0L) {
stop("'data' must have at least one row", call. = FALSE)
}
# Guard: argument matching
method <- match.arg(method)
# Guard: range check
if (!is.numeric(trim_pct) || trim_pct < 0 || trim_pct > 0.5) {
stop("'trim_pct' must be a number between 0 and 0.5, got: ", trim_pct, call. = FALSE)
}
# --- All guards passed, begin real work ---
# ...
}
R (rlang/cli — preferred for packages):
calculate_summary <- function(data, method = c("mean", "median", "trim"), trim_pct = 0.1) {
rlang::check_required(data)
if (!is.data.frame(data)) {
cli::cli_abort("{.arg data} must be a data frame, not {.cls {class(data)}}.")
}
if (nrow(data) == 0L) {
cli::cli_abort("{.arg data} must have at least one row.")
}
method <- rlang::arg_match(method)
if (!is.numeric(trim_pct) || trim_pct < 0 || trim_pct > 0.5) {
cli::cli_abort("{.arg trim_pct} must be between 0 and 0.5, not {.val {trim_pct}}.")
}
# ...
}
General (TypeScript):
function calculateSummary(data: DataFrame, method: Method, trimPct: number): Summary {
if (data.rows.length === 0) {
throw new Error(`data must have at least one row`);
}
if (trimPct < 0 || trimPct > 0.5) {
throw new RangeError(`trimPct must be between 0 and 0.5, got: ${trimPct}`);
}
// ...
}
→ Every public fn opens w/ guards rejecting invalid before side effects/computation.
If err: validation long (>15 lines of guards) → extract validate_* helper or stopifnot() for simple type assertions.
Step 3: Meaningful Err Msgs
Every msg answers 4:
- What failed — param/op
- Where — fn name/context (auto w/
cli::cli_abort) - Why — expected vs received
- How to fix — when fix non-obvious
Good:
# What + Why (expected vs. actual)
stop("'n' must be a positive integer, got: ", n, call. = FALSE)
# What + Why + How to fix
cli::cli_abort(c(
"{.arg config_path} does not exist: {.file {config_path}}",
"i" = "Create it with {.run create_config({.file {config_path}})}."
))
# What + context
cli::cli_abort(c(
"Column {.val {col_name}} not found in {.arg data}.",
"i" = "Available columns: {.val {names(data)}}"
))
Bad:
stop("Error") # What failed? No idea
stop("Invalid input") # Which input? What's wrong with it?
stop(paste("Error in step", i)) # No actionable information
→ Msgs self-documenting — dev seeing first time diagnose + fix w/o reading source.
If err: review 3 most recent bug reports. Required reading source to understand → msgs need improvement.
Step 4: Prefer stop() vs warning()
stop() (or cli::cli_abort()) when can't produce correct result. warning() only when still meaningful but caller should know.
Rule: User could silently get wrong answer → stop() not warning().
# CORRECT: stop when result would be wrong
read_config <- function(path) {
if (!file.exists(path)) {
stop("Config file not found: ", path, call. = FALSE)
}
yaml::read_yaml(path)
}
# CORRECT: warn when result is still usable
summarize_data <- function(data) {
if (any(is.na(data$value))) {
warning(sum(is.na(data$value)), " NA values dropped from 'value' column", call. = FALSE)
data <- data[!is.na(data$value), ]
}
# proceed with valid data
}
→ stop() for incorrect results; warning() for degraded-but-valid.
If err: audit existing warning() calls. Returns nonsense after → change to stop().
Step 5: Assertions for Internal Invariants
"Should never happen" → assertions. Catches programmer errs during dev:
# R: stopifnot for internal invariants
process_chunk <- function(chunk, total_size) {
stopifnot(
is.list(chunk),
length(chunk) > 0,
total_size > 0
)
# ...
}
# R: explicit assertion with context
merge_results <- function(left, right) {
if (ncol(left) != ncol(right)) {
stop("Internal error: column count mismatch (", ncol(left), " vs ", ncol(right),
"). This is a bug — please report it.", call. = FALSE)
}
# ...
}
→ Invariants asserted → bugs surface immediately at violation site, not 3 calls later cryptic.
If err: stopifnot() msgs too cryptic → switch explicit if/stop w/ context.
Step 6: Refactor Anti-Patterns
Common anti-patterns:
Anti-pattern 1: Empty tryCatch (swallowing)
# BEFORE: Error silently disappears
result <- tryCatch(
parse_data(input),
error = function(e) NULL
)
# AFTER: Log, re-throw, or return a typed error
result <- tryCatch(
parse_data(input),
error = function(e) {
cli::cli_abort("Failed to parse input: {e$message}", parent = e)
}
)
Anti-pattern 2: Defaults masking bad input
# BEFORE: Caller never knows their input was ignored
process <- function(x = 10) {
if (!is.numeric(x)) x <- 10 # silently replaces bad input
x * 2
}
# AFTER: Tell the caller about the problem
process <- function(x = 10) {
if (!is.numeric(x)) {
stop("'x' must be numeric, got ", class(x)[[1]], call. = FALSE)
}
x * 2
}
Anti-pattern 3: suppressWarnings as fix
# BEFORE: Hiding the symptom instead of fixing the cause
result <- suppressWarnings(as.numeric(user_input))
# AFTER: Validate explicitly, handle the expected case
if (!grepl("^-?\\d+\\.?\\d*$", user_input)) {
stop("Expected a number, got: '", user_input, "'", call. = FALSE)
}
result <- as.numeric(user_input)
Anti-pattern 4: Catch-all handlers
# BEFORE: Every error treated the same
tryCatch(
complex_operation(),
error = function(e) message("Something went wrong")
)
# AFTER: Handle specific conditions, let unexpected ones propagate
tryCatch(
complex_operation(),
custom_validation_error = function(e) {
cli::cli_warn("Validation issue: {e$message}")
fallback_value
}
# Unexpected errors propagate naturally
)
→ Anti-patterns replaced w/ explicit validation or specific err handling.
If err: remove tryCatch causes cascading → upstream has validation gap. Fix source not symptom.
Step 7: Validate Refactoring
Run tests to confirm err paths work:
# Verify error messages are triggered
testthat::expect_error(calculate_summary("not_a_df"), "must be a data frame")
testthat::expect_error(calculate_summary(data.frame()), "at least one row")
testthat::expect_error(calculate_summary(mtcars, trim_pct = 2), "between 0 and 0.5")
# Verify valid inputs still work
testthat::expect_no_error(calculate_summary(mtcars, method = "mean"))
# Run full test suite
Rscript -e "devtools::test()"
→ All tests pass. Err-path tests confirm bad input triggers expected msg.
If err: existing tests relied on silent failures (returning NULL on bad input) → update to expect new err.
Check
- Every public fn validates inputs before work
- Err msgs: what, where, why, how to fix
-
stop()for incorrect results -
warning()only degraded-but-valid - No empty
tryCatchswallowing - No
suppressWarnings()as validation substitute - No defaults silently masking invalid
- Invariants use
stopifnot()or explicit assertions - Err-path tests per guard
- Test suite passes post-refactor
Traps
- Validate too deep: Validate at boundaries (public API), not every internal helper. Over-validation adds noise + hurts perf.
- Msgs no context:
"Invalid input"forces caller guess. Always param name, expected type/range, actual value. - warning() when mean stop(): Returns garbage after warn → wrong silently. Use
stop(), let caller decide. - Swallow in tryCatch:
tryCatch(..., error = function(e) NULL)hides bugs. Must catch → log or re-throw w/ context. - Forget call. = FALSE: R
stop("msg")includes call by default — noisy end users. Usecall. = FALSEuser-facing.cli::cli_abort()does auto. - Validate in tests not code: Tests verify behavior not protect production. Validation in fn itself.
- Wrong R binary hybrid systems: WSL/Docker,
Rscriptmay resolve cross-platform wrapper not native R. Checkwhich Rscript && Rscript --version. Prefer native (/usr/local/bin/RscriptLinux/WSL). See Setting Up Your Environment.
→
write-testthat-tests— tests verifying err pathsreview-pull-request— review for missing validation + silent failuresreview-software-architecture— err-handling strategy system levelcreate-skill— new skills following agentskills.iosecurity-audit-codebase— security-focused review overlapping validation
Dépôt GitHub
Compétences associées
evaluating-llms-harness
TestsCette compétence Claude exécute le lm-evaluation-harness pour évaluer les modèles de langage sur plus de 60 tâches académiques standardisées telles que MMLU et GSM8K. Elle est conçue pour permettre aux développeurs de comparer la qualité des modèles, de suivre les progrès de l'entraînement ou de rapporter des résultats académiques. L'outil prend en charge différents backends, incluant les modèles HuggingFace et vLLM.
cloudflare-cron-triggers
TestsCette compétence fournit une connaissance complète pour la mise en œuvre de Déclencheurs Cron Cloudflare afin de planifier des Workers à l'aide d'expressions cron. Elle couvre la configuration de tâches périodiques, de travaux de maintenance et de flux de travail automatisés, tout en traitant des problèmes courants tels que les expressions cron non valides et les problèmes de fuseau horaire. Les développeurs peuvent l'utiliser pour configurer des gestionnaires planifiés, tester des déclencheurs cron et intégrer avec Workflows et Green Compute.
webapp-testing
TestsCette Compétence Claude fournit une boîte à outils basée sur Playwright pour tester des applications web locales via des scripts Python. Elle permet la vérification frontend, le débogage d'interface utilisateur, la capture d'écrans et la consultation des journaux, tout en gérant les cycles de vie du serveur. Utilisez-la pour les tâches d'automatisation de navigateur, mais exécutez les scripts directement plutôt que de lire leur code source pour éviter la pollution du contexte.
finishing-a-development-branch
TestsCette compétence aide les développeurs à finaliser leur travail en vérifiant que les tests passent, puis en présentant des options d'intégration structurées. Elle guide le processus de fusion, de création de PRs ou de nettoyage des branches une fois l'implémentation terminée. Utilisez-la lorsque votre code est prêt et testé pour finaliser systématiquement le cycle de développement.
