Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.agentguardian.io/llms.txt

Use this file to discover all available pages before exploring further.

The AgentGuardian blog is where the maintainers write up the engineering work behind the project. Three buckets:
  • Releases — what shipped, why it shipped, what it changes for users. The canonical entry point for any reader landing here from a release tag.
  • Deep-dives — the attacks AgentGuardian models, the standards it maps to, the trade-offs behind the swarm architecture. Written for security engineers and principal engineers, not for procurement.
  • Walkthroughs — step-by-step attack reproductions against deliberately-vulnerable demo agents, ending with the AgentGuardian report and the mitigation diff.
We do not publish marketing copy here. If a post reads like an announcement, it has been mis-filed.

Current posts

Introducing AgentGuardian

Why we built an open-source red-teaming toolkit for AI agents, what ships in v1.1, and how to run the first scan against your own agent.

How prompt injection becomes tool abuse

The three-step cascade — recon, payload delivery, tool call — that turns a single injected instruction into data exfiltration or privilege escalation. OWASP ASI 2026 and MITRE ATLAS mapping inline.

Breaking a LangGraph agent

A full reproduction against the testbench’s coding-assistant agent — attack prompts, transcript, AIVSS score, mitigation, and a clean re-run.

AgentGuardian vs other red-team tools

Factual comparison against PyRIT, garak, Promptfoo, Inspect, DeepTeam, and manual testing. What each tool optimises for and where AgentGuardian fits.

Watch the 90-second demo

The hero video — start vulnerable agent, run a scan, read the report, ship the mitigation, re-scan to zero findings — is on YouTube and linked from the README. If you only have ninety seconds, watch that first.

Contribute

The blog source lives in docs/blog/*.mdx in the agent-guardian repository. PRs with deep-dives, walkthroughs against your own agents, or reproductions of published agent-security research are welcome — see CONTRIBUTING.md.