FAQ & troubleshooting

Is the secret redaction reliable enough to trust?

It's best-effort — it catches common patterns (private keys, JWTs, AWS/GitHub/Slack tokens, KEY=secret assignments). Always review output before sharing, and keep secrets out of the tree in the first place. Disable with --no-redact only when offline and trusting the destination.

A file I wanted was skipped

It was probably ignored (defaults + .gitignore + --ignore), detected as binary (contains a null byte), larger than 512 KB, or dropped because the token budget was reached. The CLI summary and skippedFiles tell you which.

The ignore matcher didn't behave like git

It implements a common subset of .gitignore (comments, negation, anchoring, */**), not the full spec. Add explicit --ignore patterns for edge cases.

How accurate is the token count?

It's an estimate (~4 chars/token), model-agnostic and fast — meant for budgeting, not exact billing. Real tokenizers vary by model.

Can I include normally-ignored files?

Yes for extra excludes via --ignore; to force-include something your .gitignore blocks, run against a subdirectory or temporarily adjust ignore rules.

Markdown or XML — which should I use?

Markdown is friendliest for chat; XML (--format xml) is handy for RAG pipelines that parse <file> blocks.