The definition of a conflict of Interest in peer review is a circumstance that makes you "unable to make an impartial scientific judgment or evaluation." (PNAS Editorial Policies - Competing Interest). JOSS is concerned with avoiding any actual conflicts of interest, and being sufficiently transparent that we avoid the appearance of conflicts of interest as well.
As a reviewer, COIs are your present or previous association with any authors of a submission: recent (past four years) collaborators in funded research or work that is published; and lifetime for the family members, business partners, and thesis student/advisor or mentor. In addition, your recent (past year) association with the same organization of a submitter is a COI, for example, being employed at the same institution.
If you have a conflict of interest with a submission, you should disclose the specific reason to the submissions' editor. This may lead to you not being able to review the submission, but some conflicts may be recorded and then waived, and if you think you are able to make an impartial assessment of the work, you should request that the conflict be waived. For example, if you and a submitter were two of 2000 authors of a high energy physics paper but did not actually collaborate. Or if you and a submitter worked together 6 years ago, but due to delays in the publishing industry, a paper from that collaboration with both of you as authors was published 2 year ago. Or if you and a submitter are both employed by the same very large organization but in different units without any knowledge of each other.
Declaring actual, perceived, and potential conflicts of interest is required under professional ethics. If in doubt: ask the editors.