{
  "slug": "anatomy-of-agent-harness",
  "title": "The Anatomy of an Agent Harness",
  "generated": "2026-06-07T08:53:23.819Z",
  "byId": {
    "concept-agent-harness": {
      "folder": "concepts",
      "file": "concepts/concept-agent-harness.md",
      "title": "Agent Harness",
      "type": "concept",
      "tags": [
        "architecture",
        "definitions"
      ]
    },
    "concept-filesystem-primitive": {
      "folder": "concepts",
      "file": "concepts/concept-filesystem-primitive.md",
      "title": "Filesystem as a Harness Primitive",
      "type": "concept",
      "tags": [
        "state-management",
        "infrastructure"
      ]
    },
    "concept-bash-general-tool": {
      "folder": "concepts",
      "file": "concepts/concept-bash-general-tool.md",
      "title": "Bash + Code Execution as a General Purpose Tool",
      "type": "concept",
      "tags": [
        "tool-use",
        "autonomy"
      ]
    },
    "concept-context-rot": {
      "folder": "concepts",
      "file": "concepts/concept-context-rot.md",
      "title": "Context Rot",
      "type": "concept",
      "tags": [
        "context-management",
        "failure-modes"
      ]
    },
    "concept-compaction": {
      "folder": "concepts",
      "file": "concepts/concept-compaction.md",
      "title": "Context Compaction",
      "type": "concept",
      "tags": [
        "context-management",
        "middleware"
      ]
    },
    "concept-tool-call-offloading": {
      "folder": "concepts",
      "file": "concepts/concept-tool-call-offloading.md",
      "title": "Tool Call Offloading",
      "type": "concept",
      "tags": [
        "context-management",
        "tool-use"
      ]
    },
    "concept-progressive-disclosure": {
      "folder": "concepts",
      "file": "concepts/concept-progressive-disclosure.md",
      "title": "Progressive Disclosure (Skills)",
      "type": "concept",
      "tags": [
        "context-management",
        "tool-use"
      ]
    },
    "concept-ralph-loop": {
      "folder": "concepts",
      "file": "concepts/concept-ralph-loop.md",
      "title": "Ralph Loops",
      "type": "concept",
      "tags": [
        "orchestration",
        "long-horizon"
      ]
    },
    "concept-harness-model-coevolution": {
      "folder": "concepts",
      "file": "concepts/concept-harness-model-coevolution.md",
      "title": "Coupling of Model Training and Harness Design",
      "type": "concept",
      "tags": [
        "model-training",
        "evolution"
      ]
    },
    "contrarian-harness-optimization": {
      "folder": "concepts",
      "file": "concepts/contrarian-harness-optimization.md",
      "title": "Contrarian — Native Harnesses Aren't Always Best",
      "type": "contrarian",
      "tags": [
        "contrarian",
        "optimization",
        "benchmarking"
      ]
    },
    "contrarian-harness-longevity": {
      "folder": "concepts",
      "file": "concepts/contrarian-harness-longevity.md",
      "title": "Contrarian — Harness Engineering Will Survive Model Advancements",
      "type": "contrarian",
      "tags": [
        "contrarian",
        "future-trends"
      ]
    },
    "claim-agent-equation": {
      "folder": "claims",
      "file": "claims/claim-agent-equation.md",
      "title": "Agent = Model + Harness",
      "type": "claim",
      "tags": [
        "definitions",
        "architecture"
      ]
    },
    "claim-long-horizon-compounds": {
      "folder": "claims",
      "file": "claims/claim-long-horizon-compounds.md",
      "title": "Long-horizon execution requires compounding harness primitives",
      "type": "claim",
      "tags": [
        "long-horizon",
        "architecture"
      ]
    },
    "claim-harness-overfitting": {
      "folder": "claims",
      "file": "claims/claim-harness-overfitting.md",
      "title": "Post-training with a harness creates tool-logic overfitting",
      "type": "claim",
      "tags": [
        "model-training",
        "generalization"
      ]
    },
    "framework-harness-derivation": {
      "folder": "frameworks",
      "file": "frameworks/framework-harness-derivation.md",
      "title": "Working Backwards: Deriving Harness Features",
      "type": "framework",
      "tags": [
        "design-patterns",
        "methodology"
      ]
    },
    "entity-vivek-trivedy": {
      "folder": "entities",
      "file": "entities/entity-vivek-trivedy.md",
      "title": "Vivek Trivedy",
      "type": "entity",
      "tags": [
        "author",
        "person"
      ]
    },
    "entity-langchain": {
      "folder": "entities",
      "file": "entities/entity-langchain.md",
      "title": "LangChain",
      "type": "entity",
      "tags": [
        "organization"
      ]
    },
    "entity-deepagents": {
      "folder": "entities",
      "file": "entities/entity-deepagents.md",
      "title": "deepagents",
      "type": "entity",
      "tags": [
        "library",
        "harness-engineering"
      ]
    },
    "entity-claude-code": {
      "folder": "entities",
      "file": "entities/entity-claude-code.md",
      "title": "Claude Code",
      "type": "entity",
      "tags": [
        "product",
        "agent"
      ]
    },
    "entity-codex-5-3": {
      "folder": "entities",
      "file": "entities/entity-codex-5-3.md",
      "title": "Codex-5.3",
      "type": "entity",
      "tags": [
        "model",
        "documentation"
      ]
    },
    "entity-terminal-bench-2-0": {
      "folder": "entities",
      "file": "entities/entity-terminal-bench-2-0.md",
      "title": "Terminal Bench 2.0",
      "type": "entity",
      "tags": [
        "benchmark"
      ]
    },
    "entity-opus-4-6": {
      "folder": "entities",
      "file": "entities/entity-opus-4-6.md",
      "title": "Opus 4.6",
      "type": "entity",
      "tags": [
        "model"
      ]
    },
    "entity-context7": {
      "folder": "entities",
      "file": "entities/entity-context7.md",
      "title": "Context7",
      "type": "entity",
      "tags": [
        "tool",
        "mcp"
      ]
    },
    "entity-langsmith": {
      "folder": "entities",
      "file": "entities/entity-langsmith.md",
      "title": "LangSmith",
      "type": "entity",
      "tags": [
        "platform",
        "debugging"
      ]
    },
    "quote-harness-definition": {
      "folder": "quotes",
      "file": "quotes/quote-harness-definition.md",
      "title": "If you're not the model, you're the harness.",
      "type": "quote",
      "tags": [
        "definitions"
      ]
    },
    "quote-context-engineering": {
      "folder": "quotes",
      "file": "quotes/quote-context-engineering.md",
      "title": "Harnesses as context delivery",
      "type": "quote",
      "tags": [
        "context-management"
      ]
    },
    "quote-intelligence-vs-usefulness": {
      "folder": "quotes",
      "file": "quotes/quote-intelligence-vs-usefulness.md",
      "title": "Intelligence vs Usefulness",
      "type": "quote",
      "tags": [
        "architecture"
      ]
    },
    "action-implement-filesystem": {
      "folder": "action-items",
      "file": "action-items/action-implement-filesystem.md",
      "title": "Implement Filesystem Abstractions",
      "type": "action",
      "tags": [
        "infrastructure",
        "state-management"
      ]
    },
    "action-provide-bash": {
      "folder": "action-items",
      "file": "action-items/action-provide-bash.md",
      "title": "Provide Bash/Code Execution",
      "type": "action",
      "tags": [
        "tool-use",
        "autonomy"
      ]
    },
    "action-tool-call-offloading": {
      "folder": "action-items",
      "file": "action-items/action-tool-call-offloading.md",
      "title": "Implement Tool Call Offloading",
      "type": "action",
      "tags": [
        "context-management"
      ]
    },
    "action-use-ralph-loops": {
      "folder": "action-items",
      "file": "action-items/action-use-ralph-loops.md",
      "title": "Use Ralph Loops for Long-Horizon Tasks",
      "type": "action",
      "tags": [
        "orchestration",
        "long-horizon"
      ]
    },
    "prereq-react-loop": {
      "folder": "prerequisites",
      "file": "prerequisites/prereq-react-loop.md",
      "title": "ReAct Loop",
      "type": "prereq",
      "tags": [
        "agent-patterns"
      ]
    },
    "prereq-mcp": {
      "folder": "prerequisites",
      "file": "prerequisites/prereq-mcp.md",
      "title": "Model Context Protocol (MCP)",
      "type": "prereq",
      "tags": [
        "protocols",
        "tool-use"
      ]
    },
    "question-orchestrating-hundreds": {
      "folder": "open-questions",
      "file": "open-questions/question-orchestrating-hundreds.md",
      "title": "Orchestrating Hundreds of Agents",
      "type": "question",
      "tags": [
        "scale",
        "orchestration"
      ]
    },
    "question-self-analyzing-traces": {
      "folder": "open-questions",
      "file": "open-questions/question-self-analyzing-traces.md",
      "title": "Agents Analyzing Own Traces",
      "type": "question",
      "tags": [
        "self-improvement",
        "debugging"
      ]
    },
    "question-jit-tool-assembly": {
      "folder": "open-questions",
      "file": "open-questions/question-jit-tool-assembly.md",
      "title": "Just-In-Time Tool Assembly",
      "type": "question",
      "tags": [
        "tool-use",
        "context-management"
      ]
    }
  },
  "byFolder": {
    "concepts": [
      "concept-agent-harness",
      "concept-filesystem-primitive",
      "concept-bash-general-tool",
      "concept-context-rot",
      "concept-compaction",
      "concept-tool-call-offloading",
      "concept-progressive-disclosure",
      "concept-ralph-loop",
      "concept-harness-model-coevolution",
      "contrarian-harness-optimization",
      "contrarian-harness-longevity"
    ],
    "claims": [
      "claim-agent-equation",
      "claim-long-horizon-compounds",
      "claim-harness-overfitting"
    ],
    "frameworks": [
      "framework-harness-derivation"
    ],
    "entities": [
      "entity-vivek-trivedy",
      "entity-langchain",
      "entity-deepagents",
      "entity-claude-code",
      "entity-codex-5-3",
      "entity-terminal-bench-2-0",
      "entity-opus-4-6",
      "entity-context7",
      "entity-langsmith"
    ],
    "quotes": [
      "quote-harness-definition",
      "quote-context-engineering",
      "quote-intelligence-vs-usefulness"
    ],
    "action-items": [
      "action-implement-filesystem",
      "action-provide-bash",
      "action-tool-call-offloading",
      "action-use-ralph-loops"
    ],
    "prerequisites": [
      "prereq-react-loop",
      "prereq-mcp"
    ],
    "open-questions": [
      "question-orchestrating-hundreds",
      "question-self-analyzing-traces",
      "question-jit-tool-assembly"
    ]
  },
  "byTag": {
    "architecture": [
      "concept-agent-harness",
      "claim-agent-equation",
      "claim-long-horizon-compounds",
      "quote-intelligence-vs-usefulness"
    ],
    "definitions": [
      "concept-agent-harness",
      "claim-agent-equation",
      "quote-harness-definition"
    ],
    "state-management": [
      "concept-filesystem-primitive",
      "action-implement-filesystem"
    ],
    "infrastructure": [
      "concept-filesystem-primitive",
      "action-implement-filesystem"
    ],
    "tool-use": [
      "concept-bash-general-tool",
      "concept-tool-call-offloading",
      "concept-progressive-disclosure",
      "action-provide-bash",
      "prereq-mcp",
      "question-jit-tool-assembly"
    ],
    "autonomy": [
      "concept-bash-general-tool",
      "action-provide-bash"
    ],
    "context-management": [
      "concept-context-rot",
      "concept-compaction",
      "concept-tool-call-offloading",
      "concept-progressive-disclosure",
      "quote-context-engineering",
      "action-tool-call-offloading",
      "question-jit-tool-assembly"
    ],
    "failure-modes": [
      "concept-context-rot"
    ],
    "middleware": [
      "concept-compaction"
    ],
    "orchestration": [
      "concept-ralph-loop",
      "action-use-ralph-loops",
      "question-orchestrating-hundreds"
    ],
    "long-horizon": [
      "concept-ralph-loop",
      "claim-long-horizon-compounds",
      "action-use-ralph-loops"
    ],
    "model-training": [
      "concept-harness-model-coevolution",
      "claim-harness-overfitting"
    ],
    "evolution": [
      "concept-harness-model-coevolution"
    ],
    "contrarian": [
      "contrarian-harness-optimization",
      "contrarian-harness-longevity"
    ],
    "optimization": [
      "contrarian-harness-optimization"
    ],
    "benchmarking": [
      "contrarian-harness-optimization"
    ],
    "future-trends": [
      "contrarian-harness-longevity"
    ],
    "generalization": [
      "claim-harness-overfitting"
    ],
    "design-patterns": [
      "framework-harness-derivation"
    ],
    "methodology": [
      "framework-harness-derivation"
    ],
    "author": [
      "entity-vivek-trivedy"
    ],
    "person": [
      "entity-vivek-trivedy"
    ],
    "organization": [
      "entity-langchain"
    ],
    "library": [
      "entity-deepagents"
    ],
    "harness-engineering": [
      "entity-deepagents"
    ],
    "product": [
      "entity-claude-code"
    ],
    "agent": [
      "entity-claude-code"
    ],
    "model": [
      "entity-codex-5-3",
      "entity-opus-4-6"
    ],
    "documentation": [
      "entity-codex-5-3"
    ],
    "benchmark": [
      "entity-terminal-bench-2-0"
    ],
    "tool": [
      "entity-context7"
    ],
    "mcp": [
      "entity-context7"
    ],
    "platform": [
      "entity-langsmith"
    ],
    "debugging": [
      "entity-langsmith",
      "question-self-analyzing-traces"
    ],
    "agent-patterns": [
      "prereq-react-loop"
    ],
    "protocols": [
      "prereq-mcp"
    ],
    "scale": [
      "question-orchestrating-hundreds"
    ],
    "self-improvement": [
      "question-self-analyzing-traces"
    ]
  }
}