Your Agent's Config Is a Dependency With No Lockfile
Your Agent's Config Is a Dependency With No Lockfile Your dependency policy is probably decent. Locked versions, hash verification, an SBOM if you're serious, a review gate for new packages. Meanwhile your coding agent imports instructions, skills, hooks, tool declarations, and subagents from public marketplaces and repositories. It runs all of it with your privileges, in your working directory,…
Your AI coding agent imports code snippets, skills, and configuration from public sources without a lockfile, potentially introducing security risks. A study found that 16% of these setups contain security defects, such as installing MCP servers without a pinned version or pre-approving arbitrary execution through scoped permissions.
This differs from traditional package dependencies, which are executed at build or test time in a controlled environment. The AI agent config, however, runs continuously in the user's directory with privileges matching those of the user. The study, published in September 2026, analyzed 3,171 GitHub repositories and recommends pinning MCP server versions, skills, and hooks, as well as reviewing agent configurations like dependency changes.
Until the ecosystem matures, manual hygiene measures like pinning, scanning, and minimal grants are necessary to mitigate these risks.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
