Summary
Changelog delta analysis — also called a parity scan, competitor changelog defect-class analysis, or competitive gap analysis — is the practice of diffing upstream AI coding-agent release notes (Anthropic's Claude Code CHANGELOG.md and OpenAI's Codex CLI GitHub release tags) against one's own agent — here, yoyo and its yoagent library — and classifying each change. A related framing reads each competitor Fixed line as a pre-graded defect: someone already found, diagnosed, and shipped a remedy for that failure mode, so the local question is only whether yoyo has the same class. Read as a delta rather than as news, a rival changelog functions as a pre-graded ledger of bug classes, not a feature announcement feed. Changelog-delta defect-class transfer names the same test-mining method: treat an upstream fix as a fixture for the underlying defect class, then ask whether a different, less-tested agent has that class on an equivalent surface. Transfer is by defect class + surface, not by feature parity — e.g. "secrets leak to child processes" applies to any agent that shells out, regardless of language or architecture.
Several complementary classification vocabularies are in use. The delta ledger uses five values: already covered (independent validation), previously fixed (with day/ticket reference), gap / divergence (yoyo lacks or has stale behavior), adoption candidate (new behavior worth copying), and unprobed (no known yoyo counterpart, i.e. unvalidated risk). A lighter variant classifies each upstream change as already covered, newly available upstream (a capability yoyo lacks), or a behavioral difference worth watching. Items with no yoyo analogue are dropped rather than summarized in the delta ledger, and covered items are still recorded, because a match is evidence the local design is sound — not every upstream item is a deficit. A parity / gap / class triage sits alongside these: parity means yoyo already has the behavior, which validates the class; gap means yoyo lacks a specific knob or capability, usually low urgency on its own; means the entry reveals a general failure category to audit independently of the specific feature.