AI coding agents route around the repository gate, and JFrog moves the checkpoint to the agent
Agent Package Resolution, now in preview, binds Claude Code and Cursor to Artifactory. Two of its three enforcement layers still run where the agent does.
The Watch · Security desk
What happened
- AI coding agents now pick dependencies themselves from a natural-language prompt, with no package named by the user.
- JFrog put Agent Package Resolution into preview, wiring Claude Code and Cursor to resolve packages through Artifactory.
- Researchers found nearly a fifth of AI package recommendations name packages that do not exist, which attackers register.
Why it matters
- constraint Reviewing manifests and hardening developer environments no longer covers the requester that matters, because the download and the decision both happen without a human writing a dependency line.
- exposure Hallucinated package names get installed by a requester with no instinct for an unfamiliar library, so the pre-registered typo has nobody left to catch it.
- contradiction JFrog claims enforcement does not depend on agent compliance, yet two of its three layers live in the agent session and the local package manager, where compliance is exactly what is being assumed.
- decision Buyers have to decide whether a per-agent integration list is a governance model at all, or a stopgap until the control moves to network egress and registry credentials.
Start with the arithmetic in JFrog's own report. If 2025 closed with more than 171,000 unique malicious npm packages after a 451% rise [4], the year opened from a base of roughly 31,000 [9]. That is the volume the curated repository was built to filter, and the filter only works on traffic that arrives at it.
The three described layers are not equally durable. Session steering hands the agent an Artifactory URL for each governed package type at the start of a session [5], which means the agent has to choose to use it. The `jf setup` binding is stronger: it configures the local package manager per project, survives sessions and tools, and catches the installs nobody typed, including postinstall scripts and `pip install -r requirements.txt` [5][6]. Only the third layer, server-side Curation, sits somewhere the requester cannot negotiate with [5]. JFrog writes that each layer enforces independently and that governance does not rely on agent compliance [11]; by the post's own description, two of the three execute in the client environment [10].
The server-side wall governs what passes through Artifactory [5]. It says nothing about a request that never goes there. The post does not describe blocking direct access to public registries or controlling agent egress [8], and public registries remain the default destination for agent-resolved dependencies across npm, PyPI, Maven, Go, Docker Hub, Helm and NuGet [2]. So the control that matters is routing and credentials, not the manifest. Nobody wrote a manifest: the agent picked the packages from a natural-language prompt [3].
Coverage is an allowlist. The feature ships for Claude Code and Cursor, with more agents promised [1]. Every agent, wrapper and CI runner outside that list keeps the behaviour JFrog describes as default rather than exceptional, with scanning and audit trails never applied [7]. And the fix is described entirely in JFrog parts, Artifactory, Curation and Xray [1][6], which is worth remembering when reading a vendor's account of the gap.
What to watch
- Whether JFrog names the further agents and CLIs it supports, and how long unsupported tools stay outside the gate.
- Whether any vendor publishes measured bypass rates for session steering, rather than describing the layer as independent enforcement.
- Whether enterprises start blocking direct egress to public registries from agent runtimes, forcing all resolution through a proxy.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence34
- Adoption14
- Hype gap+42
- Incentives86
- Confidence44
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
JFrog announced Agent Package Resolution (in preview), a feature of the JFrog Agent Plugin that lets AI coding agents natively resolve packages through JFrog Artifactory; it is available for Claude Code or Cursor now, with more agents said to be coming soon.
- [2]
By default, dependencies retrieved by AI coding agents are pulled directly from public registries: npm, PyPI, Maven, Go, Docker Hub, Helm and NuGet.
- [3]
With tools like Cursor and Claude Code, users describe intent in natural language and the agent decides what packages are needed and retrieves them autonomously; a prompt such as building a stamp-collection app can trigger a cascade of dependency downloads without the user naming a single package.
- [4]
Malicious npm package activity surged 451% in 2025, reaching more than 171,000 unique malicious packages, according to JFrog's Software Supply Chain Security State of the Union 2026 report.
- [5]
Agent Package Resolution has three layers: session steering, which gives the agent the Artifactory URL for each governed package type at the start of each session; persistent package-manager configuration via a one-time jf setup command that binds the local package manager to an Artifactory repository for a project and persists across sessions and tools; and server-side JFrog Curation enforcement, which governs what is allowed through Artifactory regardless of the other two layers.
- [6]
Coverage claimed includes direct installs resolving from governed Artifactory repositories, indirect and transitive installs such as postinstall scripts and bulk pulls from pip install -r or npm ci, JFrog Curation policy blocking malicious packages and unapproved licenses at the gate, and JFrog Xray scanning every artifact for known CVEs, license violations and operational risk.
- [7]
JFrog says agent-driven downloads happen outside controlled repositories, so curation policies, security scanning and audit trails are never applied, and that this is the default behaviour rather than an edge case.
- [8]
The JFrog post describes routing agent package requests through Artifactory and enforcing policy there; it does not describe blocking direct network access to public registries or controlling the agent's egress path.
- [9]
A 451% rise ending above 171,000 unique malicious npm packages implies a starting base of roughly 31,000 packages before the increase.
- [10]
Of the three enforcement layers JFrog describes, two run in the client environment (the agent session and the developer's local package manager) and one runs server-side.
- [11]
JFrog states that each layer provides independent enforcement and that governance does not rely on agent compliance but is enforced end to end.
- [12]
Researchers documented nearly 20% of AI-generated package recommendations referencing packages that do not exist, enabling attacks researchers call slopsquatting, in which attackers register malicious packages under names AI models repeatedly hallucinate.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- jfrog.comCoding Agents Just Reopened Your Software Supply Chain Blind Spot
1 article · August 25, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- Dependency Governance and Artifact RepositoriesFollow
- Software Supply Chain SecurityFollow
- AI Coding AgentsFollow
- Package Hallucination and SlopsquattingFollow
- Agent Guardrails and EnforcementFollow
Entities
- JFrogFollow
- Agent Package ResolutionFollow
- JFrog Agent PluginFollow
- ArtifactoryFollow
- JFrog CurationFollow
- JFrog XrayFollow
- Claude CodeFollow
- CursorFollow
- npm registryFollow
- PyPIFollow
- MavenFollow
- Docker HubFollow
- HelmFollow
- NuGetFollow
- GoFollow
- SlopsquattingFollow
- JFrog Software Supply Chain Security State of the Union 2026Follow