{"path":"AGENT_EVAL.md","size":1422,"locked":false,"binary":false,"truncated":false,"content":"# Agent Evaluation Metadata: Agent Workspace Guard SDK\n\n```json\n{\n  \"product_name\": \"Agent Workspace Guard SDK\",\n  \"sku\": \"MIMARIO-01-SAFEWORK\",\n  \"version\": \"1.0.0\",\n  \"language\": \"JavaScript (ESM)\",\n  \"runtime\": \"Node.js >= 18.0.0\",\n  \"runtime_dependencies\": [],\n  \"dev_dependencies\": [],\n  \"verified_capabilities\": [\n    \"Strict path containment preventing relative and absolute directory escape\",\n    \"Anti-symlink traversal protection via lstat and POSIX O_NOFOLLOW\",\n    \"Blacklist filtering of sensitive credentials (.env, .npmrc, id_rsa, cloud credentials)\",\n    \"Command policy evaluation blocking destructive patterns (sudo, rm -rf, chmod 777, fork bombs)\",\n    \"Shell chaining prevention blocking semicolons, pipes, logical operators, and subshells\",\n    \"Continuous SHA-256 cryptographic audit chaining of all tool actions\"\n  ],\n  \"known_limitations\": [\n    \"In-process guardrail library; does not provide OS kernel hypervisor isolation\",\n    \"Pre-compiled native binaries invoked by the agent can bypass in-process checks if allowed\",\n    \"Audit log timestamp integrity depends on host system clock\"\n  ],\n  \"installation\": \"cp -r src/ your_project/src/agent-guard/\",\n  \"test_command\": \"node --test tests/adversarial-red-team.test.mjs\",\n  \"example_command\": \"node examples/agent-runner-example.mjs\",\n  \"integration_estimate_minutes\": 10,\n  \"license\": \"Genesis Conservative Commercial Source License 1.0\"\n}\n```\n"}