Spreadsheet. The first sheet contains, by column:

• The word itself, in citation form. If the conlang uses Latin/Cyrillic/Greek I keep it in the original, otherwise I romanise the writing system, as close to possible to the original*. No idea how I’d deal with a logographic system.
• Phonemic and phonetic transcriptions. Often more than one if I’m handling multiple con-dialects.
• Part of speech: noun, verb, etc.
• Declension/conjugation, if applicable. Heavily irregular words get marked “irregular” here.
• Sandhi classes, gender, stuff like this, when applicable. Basically, ways this word interferes on other words.
• Coarse translation, with one or two words.
• Notes for subtle irregularities (e.g. “genitive in -n”), and a finer translation than the above.

The next sheets include declension/conjugation tables, separated by part of speech. Heavily irregular words get handled here, as their own conjugations/declensions.

*even if this means doing counter-intuitive stuff, like writing ⟨b̃⟩ for /m/ because it’s how the native script handles it.