
AI
Prompt injection defense for agents: what actually works
Prompt injection defense for agents that holds: deny rules, hooks that exit 2, sandboxed shell and MCP approval, each with the limits its docs spell out.
7 min read
Tag
55 posts · Page 4 of 5

AI
Prompt injection defense for agents that holds: deny rules, hooks that exit 2, sandboxed shell and MCP approval, each with the limits its docs spell out.
7 min read

AI
Cap AI coding agent spend per developer with per-member limits, gateway caps by day, week or month, Copilot per-user budgets and a per-run dollar flag.
7 min read

AI
The main risks of AI coding agents, from prompt injection to leaked secrets and unvetted MCP servers, each paired with a control and the point where it stops.
5 min read

AI
An MCP gateway routes and logs tool traffic. Policy enforcement decides allow, deny or hold on each call. Here is how they differ and which one can stop a call.
6 min read

AI
Stop PII and national IDs reaching LLMs: detect them with checksums and NER, then redact, hold or block each match in Claude Code hooks or at a gateway.
8 min read

AI
Audit Claude Code with OpenTelemetry tool_decision and tool_result events, plus hooks that keep a local copy. Refusals show in tool_decision, never tool_result.
5 min read

AI
Track Claude Code cost per developer with OpenTelemetry identity and contracted rates. Copilot shows cost per developer in its dashboard. Budgets set the cap.
9 min read

AI
Rebuild what an AI agent did from logs with trusted timestamps, and detect later edits with digests kept away from the logs they cover, not beside them.
7 min read

AI
Block Claude Code from reading .env and secret files with Read deny rules, a PreToolUse hook and sandboxing, and see which reads each layer misses, and why.
6 min read

AI
Find MCP servers in repos with .mcp.json and claude mcp list, then in clusters by listing pods and images. Covers scopes, approvals and shadow servers.
6 min read

AI
Audit MCP tool calls in Claude Code with OpenTelemetry tool events and OTEL_LOG_TOOL_DETAILS=1, plus hooks that record each server's source and denied calls.
6 min read

AI
Claude Code managed settings sit above every developer file. Deploy managed-settings.json per OS or through MDM, check it with /status, then lock the rules.
7 min read