Adversarial red/blue LLM testing via @metaharness/redblue — generates attacks across OWASP LLM Top-10 / NIST AI RMF families (prompt injection, tool misuse, data leakage, jailbreaks, denial-of-wallet), runs them against an LLM target YOU OWN, judges compromise, optionally applies declarative blue...
AI agents invoke metaharness_redblue to trigger actions in Ruflo. What it does depends on the arguments the agent supplies, and its effects often reach beyond the immediate call — builds kicked off, notifications sent, workflows started.
This tool executes adversarial attack workflows against LLM systems. While framed as a security testing utility, it triggers code execution (attack generation and injection), runs external operations (testing against target systems), and applies modifications (blue-team patches). The effects are contingent on which target is specified and which attacks/patches are selected, making it Execute rather than Read.
From the tool's definition Tool generates attacks ('generates attacks across OWASP LLM Top-10'), runs them against targets ('runs them against an LLM target YOU OWN'), executes test payloads including prompt injection, jailbreaks, denial-of-wallet attacks, and applies patches…
Attacks that exploit this kind of access
Adversarial red/blue LLM testing via @metaharness/redblue — generates attacks across OWASP LLM Top-10 / NIST AI RMF families (prompt injection, tool misuse, data leakage, jailbreaks, denial-of-wallet), runs them against an LLM target YOU OWN, judges compromise, optionally applies declarative blue-team patches, retests, and emits a board-readable report with measured failure reduction. Use when shipping an LLM-powered product and you need a repeatable security gate before exposing it to users — eyeballing prompts is wrong because attack surface coverage requires the OWASP/NIST taxonomy and the judge has to be model-driven for jailbreak detection. SAFETY: upstream hard-enforces no-creds / no-live-targets / no-shell / no-network / no-eval at config-load time; cannot be relaxed via flags. For CI / offline use --mockJudge=true ($0 marker fixture). For real model judging set $OPENROUTER_API_KEY and accept the per-run cost capped by max_cost_usd (default $3). It is categorised as a Execute tool in the Ruflo MCP Server, which means it can trigger actions or run processes. Use rate limits and argument validation.
Register the Ruflo MCP server in PolicyLayer and add a rule for metaharness_redblue: allow, deny, rate-limit, or require approval. Point your MCP client at the PolicyLayer proxy URL and the rule is enforced on every call, before it reaches Ruflo. Nothing to install.
metaharness_redblue is a Execute tool with high risk. Execute tools should be rate-limited and have argument validation enabled.
Yes. Add a rate_limit block to the metaharness_redblue rule in your PolicyLayer policy. For example, setting max: 10 and window: 60 limits the tool to 10 calls per minute. Rate limits are tracked per agent session and reset automatically.
Set action: deny in the PolicyLayer policy for metaharness_redblue. The AI agent will receive a policy violation error and cannot call the tool. You can also include a reason field to explain why the tool is blocked.
metaharness_redblue is provided by the Ruflo MCP server (ruvnet/ruflo). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
metaharness_redblue is one line of Ruflo's registry record.
The record carries the whole server: verified identity, auth posture, risk grade, every tool classified, recommended policy — re-checked continuously.
Teams ship this data inside their own products. See what a licence covers →