The latest release recorded for prismorsec/prismor is v1.68.3. However, the provided materials contain no changelog for this version. The repository is a self-hosted runtime control layer that observes AI agent tool calls before execution and helps people approve or block dangerous calls. It is written in Python and released under the Apache-2.0 license.
This tool is used to inspect and control calls related to secret exposure, prompt injection, and supply chain issues when running agents in your own operational environment. HITL (human-in-the-loop) means that a human takes part in the approval process. Prismor describes features for reviewing agent activity in a local dashboard and approving or blocking calls according to policy. The repository cites Claude, codex, langchain and others as integration examples, and provides guidance for multiple agent runtime environments along with a scope table, security check features, and usage procedures.
What can be confirmed from the materials is the tool's description and usage guidance; no newly changed features are given for v1.68.3. Therefore, there is insufficient basis to pin down the update contents of this version or to evaluate its blocking effectiveness. The design of reviewing and controlling agent calls before execution is confirmed, but the actual scope and effects cannot be judged from the repository's guidance alone.