CLI reference

Usage

scruple [patterns...] [options]

The optional check alias is retained for compatibility.

Options

Option	Description
--cache-dir <path>	Set the decision cache directory
-c, --config <path>	Use an explicit configuration file
--explain	Include accepted and abstained decisions
-f, --format <format>	stylish or json; defaults to stylish
-h, --help	Print usage information
--no-cache	Run without reading or writing cached provider responses

File patterns

Positional patterns are globbed from the current working directory. Without positional patterns, the CLI uses config.include or the default **/*.{js,jsx,mjs,cjs,ts,tsx,mts,cts}.

Dependencies, dist, build, coverage output, and .git are always ignored. config.ignore adds more patterns.

Decision cache

By default, the CLI stores successful provider responses in node_modules/.cache/scruple. Set config.cache to another DecisionCache strategy for remote, multi-tier, or other storage, or to false to disable caching. The key includes the provider ID and the complete request, so it covers both collection classifications and final decisions. Changing the evidence, question, answer labels, or provider ID produces a miss. Changing only a rule's warning threshold or message reuses the existing answer.

Cache hits do not make provider requests or add tokens to the current run. JSON statistics report cacheHits separately from requests. Invalid entries are discarded, and cache read or write failures do not stop the check.

Use --no-cache when checking live provider behavior. Use --cache-dir to replace either the default or configured strategy with a filesystem cache for that invocation. Delete the cache directory to clear all saved decisions. Pin moving model aliases when results must remain tied to one model version.

Configuration discovery

Without --config, Scruple checks these names in order:

1. scruple.config.ts
2. scruple.config.mts
3. scruple.config.js
4. scruple.config.mjs

Exit codes

Code	Meaning
0	No error-severity findings and no operational errors
1	At least one error-severity finding
2	Configuration, parsing, or provider failure

JSON output

--format json prints the complete run result, including diagnostics, operational errors, and statistics. Use it for CI integrations that need structured locations, probabilities, confidence, model IDs, cache hits, or token counts.

Add --explain to retain every final candidate decision in a decisions array, including safe answers and insufficient_context abstentions that do not produce diagnostics. Collection classifications are filtering steps and are not included. Without --explain, final answers that do not diagnose are discarded to keep normal output small. Stylish explain output prints one compact line per evaluated candidate.