← ClaudeAtlas

defend-ai-agentslisted

Securing and monitoring tool-calling AI agents against excessive agency and indirect prompt injection, where one hostile instruction becomes a real action such as sending email, moving money, or exfiltrating data (MITRE ATLAS AML.T0051, AML.T0051.001, AML.T0053; MITRE ATT&CK T1698, T1700, T1059, T1552.001, T1105; OWASP LLM06). Keys off agent tool-call telemetry (tool calls, tool-result provenance, approvals) and, for coding agents, endpoint process telemetry. Ships three behavioural detections over tool-call logs, two Sigma rules and a Defender XDR hunt for hijacked AI coding agents (CLI started headless with approvals off, risky command in the agent's shell or VS Code terminal), a least-privilege containment architecture, framework mappings, and a hunt. Use when reviewing an agent's tool scopes, vetting MCP servers, containing prompt injection, writing detections over agent tool-call or endpoint logs, triaging an agent that took an unexpected action, or hardening against the confused-deputy problem.
umbrasec-labs/umbra-blue · ★ 0 · AI & Automation · score 62
Install: claude install-skill umbrasec-labs/umbra-blue
# Securing Tool-Calling AI Agents - Detection and Containment A chatbot that gets prompt-injected says something wrong. An agent that gets prompt-injected does something wrong: it sends the email, runs the query, moves the money. The moment you give a model tools, its output stops being text and becomes actions, and the security boundary moves from "what can it say" to "what can it do." This skill is defensive: it describes the agent attack surface only as far as a defender needs to log, detect, and contain it, and ships no injection payloads, jailbreak strings, or evasion techniques. ## TL;DR for defenders An agent is an LLM in a loop with tools, and its tool calls are where impact lives. Treat every tool call as a privileged action, not a chat message. You cannot patch prompt injection, so design for the assumption that the agent will eventually follow a hostile instruction and limit what that costs. The four controls that matter most: **scope every tool to the task** (least privilege beats excessive agency), **put a human in front of consequential, irreversible actions**, **enforce policy on tool calls outside the model** (a system prompt is guidance, not a boundary), and **give the agent its own scoped identity** rather than the user's full token. Log every tool call - that is your process-creation telemetry. The highest-precision detection below is rule 1, the out-of-profile tool call against a tight per-agent capability profile; rule 2, a consequential action right a