Prismor Adds Policy Checks and Blocking for AI Coding Agent Actions
Summary
Prismor is an open-source layer that checks AI coding agents’ tool calls before execution. Its features include configurable blocking, package-install checks, and an optional semantic prompt-injection guard; performance and detection figures are vendor-reported.
Key points
- Prismor checks AI coding agents’ tool calls against policy before execution, returning allow, warn, or block.
- It logs actions in observe mode by default; users can enable rules to block actions in real time.
- Package checks score installs using package age, maintainer count, install scripts, and known indicators; scores of 60 or higher are blocked, as are indicator matches.
- An optional semantic guard sends uncertain prompt-injection cases to an LLM for review, according to PrismorSec.
- PrismorSec reports a 30% improvement in detection across more than 800 test cases, with no increase in harmless-text flags.
- The vendor reports 0.8 milliseconds of added delay per tool call across 10,000 simulated sessions.
Article Details
- Event Type
- Release of an open-source security layer for AI coding agents
- Impact
- Prismor checks AI-agent tool calls against policy before execution and can allow, warn, or block them. It also logs activity, scores package installs, and offers an optional semantic guard for prompt-injection detection. PrismorSec reports a 0.8-millisecond average added delay per tool call across 10,000 simulated sessions and a 30% improvement in prompt-injection catch rate in tests of more than 800 cases.
Vendors
Products
Claude CodeIt sits between an agent such as Claude Code, Codex, or Cursor and the actions that agent wants to take, and it checks each tool call against a policy before the call runs.Codex#PrismorCursor#PrismorPrismorPrismor is a free, open-source security layer for AI coding agents that checks each tool call against policy before it runs, helping stop risky actions like credential leaks, poisoned file redirections, and compromised