metaharness_security_bench
ADR-153 — upstream Darwin Shield (their own ADR-155): evolves a champion security-detection harness against a 10-vuln/9-decoy ground-truth corpus and grades on TPR/FPR/patch-pass/repro/unsafe vs four baselines (B0 static, B1 LLM-single-pass, B2 fixed-agent, B3 Darwin-champion). Closest reference ...
This record as markdown: /tools/io-github-ruvnet-claude-flow/metaharness-security-bench.md
What metaharness_security_bench does on Claude Flow
AI agents invoke metaharness_security_bench to trigger actions in Claude Flow. 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.
Why metaharness_security_bench is rated High
This tool executes an automated security benchmarking harness that runs vulnerability detection, patch validation, and reproduction tests against a ground-truth corpus. It actively runs code and external operations (evolving a champion harness, grading patch-pass/repro/unsafe outcomes) rather than merely reading data.
From the tool's definition 'evolves a champion security-detection harness against a 10-vuln/9-decoy ground-truth corpus and grades on TPR/FPR/patch-pass/repro/unsafe vs four baselines' and 'running this periodically gives baseline diversity and week-over-week champion-fitness drift'
Attacks that exploit this kind of access
The rule that runs metaharness_security_bench safely
PolicyLayer is an MCP gateway: it sits between your AI agents and Claude Flow, and checks every tool call against a rule you set before the call runs. Nothing changes on the server itself. For metaharness_security_bench, this is the rule to start with:
metaharness_security_bench stays usable, but rate-capped: a runaway agent can't fire it dozens of times a minute. Everything else on the server is denied unless you say otherwise.
The button opens the PolicyLayer dashboard: create your workspace, connect Claude Flow, apply this rule, and every metaharness_security_bench call is checked against it from then on.
Questions about metaharness_security_bench
ADR-153 — upstream Darwin Shield (their own ADR-155): evolves a champion security-detection harness against a 10-vuln/9-decoy ground-truth corpus and grades on TPR/FPR/patch-pass/repro/unsafe vs four baselines (B0 static, B1 LLM-single-pass, B2 fixed-agent, B3 Darwin-champion). Closest reference implementation for ruflo ADR-155 nightly self-learning security harness (#2417). Use when you need an empirical floor for Loop A reward-signal soundness; running this periodically gives baseline diversity and week-over-week champion-fitness drift. Bypassing this and just running the static MCP scan is wrong because static-only baseline (B0) reaches TPR=0.3/FPR=1 — proving static-alone has a measured detection ceiling. Parses overall PASS/FAIL + per-gate verdicts + baselines table from markdown. It is categorised as a Execute tool in the Claude Flow MCP Server, which means it can trigger actions or run processes. Use rate limits and argument validation.
Register the Claude Flow MCP server in PolicyLayer and add a rule for metaharness_security_bench: 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 Claude Flow. Nothing to install.
metaharness_security_bench 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_security_bench 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_security_bench. 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_security_bench is provided by the Claude Flow MCP server (claude-flow). PolicyLayer sits as a proxy in front of this server to enforce policies before tool calls reach the server.
More on Claude Flow, and thousands of servers like it.
Across the catalogue