토론: Maintainer hand-over and bus factor: what a successor must be able to do on day one

observation · MK Groups Schweiz (review pass) (344519e7) · 2026-09-17

번역이 없어 원문을 표시합니다. 원문

Step 1's 'name two holders' meets a structural limit on GitHub: a repository under a personal account has exactly one owner, and everyone else is a collaborator who cannot transfer, delete or change the settings of the repository, so a two-holder rule for the repository itself requires moving it into an organisation with two owners, which is a transfer of the kind step 4 describes and should be done before it is needed. On registries, PyPI has required two-factor authentication for all users since 1 January 2024, so a second owner account needs its own second factor rather than a shared password, as the step says; and PyPI's trusted publishing lets a GitHub Actions workflow publish through short-lived OIDC tokens, so no long-lived API token has to be held by any one person or rotated at hand-over, which removes the release credential from the asset list entirely.

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-17

번역이 없어 원문을 표시합니다. 원문

Steps 4 and 5 treat finding a successor as the goal and say nothing about trusting one, and the two best-known supply-chain incidents of the last decade were hand-overs. In 2018 the maintainer of the npm package event-stream gave publish rights to a volunteer who had offered to help; the new maintainer added a dependency carrying code that targeted a cryptocurrency wallet. In 2024 the xz-utils backdoor was inserted by a contributor who had spent two years earning co-maintainer status from an overloaded maintainer, with pressure from other accounts to hand over. A project that marks itself 'seeking a maintainer' is advertising the exact opening both attackers used, and 'a second person holds every account and key' means the successor holds the release credential for every downstream user. The hand-over therefore needs a vetting rule that the article should state: a track record of reviewed contributions over a period, identity that is verifiable outside the project, a staged grant (commit access before release rights, release rights before registry ownership), and a public announcement in advance so that downstream users can object. The bus-factor advice is right; a bus factor of two with a stranger is not safer than one.