RepoRadar
Leave a Tip
Item detail

msoedov/agentic_security

Agentic Security is an Apache-2.0 agentic LLM vulnerability scanner and AI red-teaming kit from msoedov. 1,904 stars, ships a curated library of multi-step attack scenarios (data exfiltration, prompt-injection chains, indirect-prompt-injection via documents, jailbreak cascades) plus a programmatic API for running them against any agent surface. Designed to integrate with CI pipelines and report a

Open source →Back to radar
Score7.5
Popularity78.0
Riskconditional
TierSilver
Score breakdown
Usefulness8.0
Novelty7.0
Momentum8.0
Maturity7.1
Open-source/build8.4
Evidence7.2
Workflow potential8.6
Setup ease6.4

Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.

Why it matters

Useful for security and platform teams that need to red-team agents before shipping them. Point Agentic Security at an agent endpoint, run the bundled attack scenarios, and use the per-class failure report to decide which agent workflows need additional guardrails before they reach production.

Who should use it

security teams red-teaming agents before promoting them to productionplatform teams wiring agent red-team checks into CI alongside unit and integration testsagent developers who need a per-attack-class failure rate (data exfil, indirect injection, tool abuse, jailbreak cascade)compliance teams building evidence for AI-risk frameworks that require documented adversarial testingresearchers studying multi-step prompt-injection attacks and how they cascade across agent tool calls

Who should skip it

Skip if the source link, docs, or setup requirements do not match your workflow.

Risk explanation

adversarial testing tool — run only against development endpoints and agent surfaces you own or have explicit authorization to test; scans for prompt-injection vulnerabilities; treat the output as a list of attack vectors to mitigate, not as a checklist of exploits.

Evidence links

Closest alternatives / related signals

agent-securityred-teamvulnerability-scannerprompt-injectiondata-exfiltrationjailbreakevaluationci