[KongchangAI]
ConceptAI依赖漂移 / 模型漂移 / 行为漂移

Dependency Drift

A phenomenon in AI Agent systems where underlying dependencies—such as model APIs, tool libraries, or vector databases—undergo behavioral changes without explicit version-change notifications, resulting in degraded output quality or functional anomalies in the system.

Timeline (last 90 days)

Oct 29

Dependency drift refers to the appearance of multiple incompatible or out-of-sync versions of the same package in a project's dependency tree, which may lead to behavioral differences that are hard to reproduce at runtime.

Unverified50%
Oct 29

For wrapper-layer packages like @ai-sdk/valibot, dependency drift may cause two sets of validation logic to coexist in the same process, and the problem usually surfaces at runtime rather than at install time.

Unverified50%
Oct 29

Common ways to converge dependency drift include: using npm ls or pnpm why to check version uniqueness, inspecting the lockfile to confirm resolution results, and forcing a unified version via overrides (npm/pnpm) or resolutions (Yarn).

Unverified50%
Sep 29

Dependency drift refers to the phenomenon where actual installed versions are inconsistent across different environments or at different times due to loosely constrained version ranges in a project.

Unverified50%
Sep 29

A common approach to preventing dependency drift is to include lock files in version control and use npm ci instead of npm install in CI/CD pipelines.

Unverified50%

All Facts (5)

Source Articles