Network-AI FileJobStore path traversal lets attacker-controlled job id read or delete arbitrary .json files

This is one of a cluster of 2026-11-02 advisories in the network-ai package: the FileJobStore component joins an attacker-supplied job id into a path with no containment check, so traversal in the id yields arbitrary .json file read or deletion relative to the processes' filesystem authority.

In an AI agent orchestration/testing deployment a job id may be sourced from external requests or agent data; if so, deleting or exposing the store's JSON state (including security/authorization state) becomes an integrity or confidentiality breach of the agent runtime.

FileJobStore constructs the on-disk job path by joining an attacker-supplied id with a .json suffix without resolving/containing it, so traversal sequences in the id escape the job store directory and let an attacker read or delete arbitrary .json files on the host.

network-ai npm package, FileJobStore/JobQueue, 5.13.4 through 5.15.4; fixed in 5.15.5

Upgrade to network-ai >= 5.15.5; validate/reject traversal sequences and enforce containment of job ids to the job store directory.