CLI reference

CLI reference

intl-ai is a single binary. Every command accepts --format json for machine-readable output, and - as --config reads TOML from stdin.

intl-ai init

Scaffold intl-ai.toml for the current project.

intl-ai init --locale-dir locales --source en --targets es,fr

intl-ai fill

Translate missing keys. Additive by default: it never overwrites existing values unless --regenerate or --stale say so.

intl-ai fill                          # all missing keys, all configured targets
intl-ai fill --locale es              # one locale (repeatable or comma-separated)
intl-ai fill --keys 'auth.*'          # glob scope (repeatable or comma-separated)
intl-ai fill --keys-file keys.txt     # one key glob per line
intl-ai fill --stale                  # re-fill AI-owned keys whose source changed
intl-ai fill --regenerate --keys 'x'  # rewrite AI-owned existing values
intl-ai fill --regenerate --yes       # confirm a whole-locale rewrite
intl-ai fill --regenerate --include-human --keys 'x'  # also human-owned values
intl-ai fill --dry-run                # report without writing anything
intl-ai fill --validate icu,judge     # run the fill gate with these checks
intl-ai fill --no-validate            # skip the configured [fill].validate gate
intl-ai fill --judge-threshold 0.9    # override the judge threshold for this run
intl-ai fill --no-cache               # skip the stat cache for this run
intl-ai fill --progress               # stream per-batch progress lines to stderr

--regenerate is destructive: it requires a --keys scope, --keys-file, or --yes to confirm a whole-locale rewrite.

intl-ai check

Report findings without writing. Exit status follows --fail-on.

intl-ai check                          # all configured checks and targets
intl-ai check --fail-on missing,invalid
intl-ai check --fail-on none           # report only, never fail
intl-ai check --locale es --keys 'auth.*'
intl-ai check --origin ai              # scope stale/modified/unreviewed to one origin
intl-ai check --self-test              # run each spec check's fixtures
intl-ai check --no-cache               # skip stat cache and the findings cache
intl-ai check --progress               # stream per-locale findings to stderr

Findings are cached incrementally in .intl-ai/check-cache.json: unchanged keys replay their previous findings (cached: true in JSON output) instead of re-running the check.

intl-ai mark

Set the origin on lockfile entries (the escape hatch for the positional review rule).

intl-ai mark 'checkout.*' --locale es --origin ai      # rebaseline as AI-owned
intl-ai mark 'checkout.*' --locale es --origin human   # claim as human-owned
intl-ai mark 'legal.*' --locale es --absent            # tombstone: never translate
intl-ai mark 'legal.*' --locale es --present           # clear a tombstone

--origin ai requires an existing entry; --origin human upserts one. Tombstoned keys are skipped by fill and not reported missing by check.

intl-ai review / intl-ai unreview

Set reviewed on lockfile entries.

intl-ai review 'checkout.*' --locale es
intl-ai unreview 'auth.*' --locale es

intl-ai status

Per-locale inventory counts. Read-only and cheap enough for agents to call before deciding what to do.

intl-ai status
intl-ai status --locale es

intl-ai lockfile

Lockfile maintenance over the intl-ai.lock.d/ shards.

intl-ai lockfile check   # validate every shard
intl-ai lockfile fmt     # rewrite shards to canonical form (idempotent)
intl-ai lockfile merge   # resolve merge conflicts in shards

intl-ai config

intl-ai config validate   # resolve and validate the config (secrets masked)
intl-ai config schema     # print the JSON Schema for the config contract

The committed schema lives at docs/public/schema/intl-ai.schema.json and is served at https://intl-ai.illo.fyi/schema/v1.json.

Global flags