Agent Harness Mastery
Course map
Level 0Agent Harness OrientationFree

What is an Agent Harness?

Understand the job boundary between prompt user, application developer, and Agent engineer.

Read

What is an Agent Harness?

"What is an Agent Harness?" means turning separate Agent Harness from prompts, chat wrappers, and ordinary application code into an executable, observable, recoverable, and acceptable Harness unit. The direct output is not a prompt. It is five-layer Harness definition card. The goal is to turn the concept into boundary, architecture map, engineering objects, code comparison, and acceptance evidence.

DefinitionWhat it is

"What is an Agent Harness?" means turning separate Agent Harness from prompts, chat wrappers, and ordinary application code into an executable, observable, recoverable, and acceptable Harness unit. The direct output is not a prompt. It is five-layer Harness definition card.

BoundaryRole ownership

It does not make the model "think harder" and it does not hide every behavior inside a chat wrapper. It defines Role ownership so Learning goal to engineering task, Agent Harness control plane, and Role matrix, deliverable list, and acceptance evidence can be traced.

CounterexampleMixing prompts, app UI, and agent runtime into one vague idea

If the implementation causes "Mixing prompts, app UI, and agent runtime into one vague idea", the boundary is not established. Go back to input contract, policy, or runtime and split it again.

Acceptancefive-layer Harness definition card

Done means delivering five-layer Harness definition card, preserving Role matrix, deliverable list, and acceptance evidence, and reviewing it through Can explain who owns context, tools, policy, and delivery.

Think of it this wayIt does not make the model "think harder" and it does not hide every behavior inside a chat wrapper. It defines Role ownership so Learning goal to engineering task, Agent Harness control plane, and Role matrix, deliverable list, and acceptance evidence can be traced. The programmer shift in this lesson is from "I can call a model or tool" to "I can design Agent Harness control plane and judge completion through Can explain who owns context, tools, policy, and delivery."
Harness definition architecture map
Rendering Mermaid...
Mermaid source
flowchart LR
  input["Input Contract"] --> boundary["Role ownership"]
  boundary --> route["Learning goal to engineering task"]
  route --> runtime["Agent Harness control plane"]
  runtime --> policy{"Policy"}
  policy -->|allow| tool["Model and Tools"]
  policy -->|block| blocked["Mixing prompts, app UI, and agent runtime into one vague idea"]
  tool --> state["State and Events"]
  runtime --> state
  state --> evidence["Role matrix, deliverable list, and acceptance evidence"]
  evidence --> artifact["five-layer Harness definition card"]
  artifact --> eval["Acceptance"]

  classDef inputNode fill:#eef4ff,stroke:#4664ff,color:#0b1220;
  classDef runtimeNode fill:#ecfdf5,stroke:#10b981,color:#0b1220;
  classDef gate fill:#fff7ed,stroke:#f59e0b,color:#0b1220;
  classDef stop fill:#fee2e2,stroke:#ef4444,color:#0b1220;

  class input,boundary,route inputNode;
  class runtime,tool,state,evidence,artifact,eval runtimeNode;
  class policy gate;
  class blocked stop;
Hermes / Pi / Fiitx relation
FrameworkUseful designHow this lesson uses it
HermesEntry points, AIAgent, tool dispatch, Session Storage, Gateway/CronJudge whether Learning goal to engineering task has long-running and delivery boundaries
Pi Agent CoreAgentSession, event stream, resource loader, compaction, save pointImplement Agent Harness control plane and a recoverable turn lifecycle
FiitxWorkbench, progress stream, approval UI, model routing, history replayExpose Role matrix, deliverable list, and acceptance evidence as a visible control plane
EvalDone criteria, trace, failure case, regressionProve five-layer Harness definition card meets acceptance criteria
Code comparisonFrom LLM call to Harness run

A plain LLM call returns an answer. A Harness run carries goal, boundary, tools, policy, and acceptance as parameters.

// Plain LLM call: returns text without boundary, state, or acceptance
const answer = await model.generate({
  prompt: "Implement: What is an Agent Harness?"
});

// Agent Harness 写法:把目标、工具、策略、状态和验收写进运行合同
const result = await harness.runHarnessOrientation1HarnessDefinition({
  lesson: "0.1 What is an Agent Harness?",
  goal: "Implement: What is an Agent Harness?",
  boundary: "Role ownership",
  runtime: "Agent Harness control plane",
  route: "Learning goal to engineering task",
  tools: runtime.tools.forBoundary("Role ownership"),
  policy: {
    watch: "Mixing prompts, app UI, and agent runtime into one vague idea",
    requireEvidence: true
  },
  eval: [
    "trace_complete",
    "artifact_exists",
    "policy_checked",
    "harness_orientation_1_Harness_definition_accepted"
  ]
});

assert(result.artifact === "five-layer Harness definition card");
assert(result.evidence.includes("Role matrix, deliverable list, and acceptance evidence"));

Key points

  1. 1

    Definition: "What is an Agent Harness?" means turning separate Agent Harness from prompts, chat wrappers, and ordinary application code into an executable, observable, recoverable, and acceptable Harness unit. The direct output is not a prompt. It is five-layer Harness definition card.

  2. 2

    Boundary: It does not make the model "think harder" and it does not hide every behavior inside a chat wrapper. It defines Role ownership so Learning goal to engineering task, Agent Harness control plane, and Role matrix, deliverable list, and acceptance evidence can be traced.

  3. 3

    Code: a plain call gets answer; a Harness call must return artifact, evidence, and trace.

  4. 4

    Acceptance: deliver five-layer Harness definition card, preserve Role matrix, deliverable list, and acceptance evidence, and guard against "Mixing prompts, app UI, and agent runtime into one vague idea".