diff --git a/command-snapshot.json b/command-snapshot.json
index 935c24d9..25d953e5 100644
--- a/command-snapshot.json
+++ b/command-snapshot.json
@@ -287,6 +287,57 @@
     ],
     "plugin": "@salesforce/plugin-agent"
   },
+  {
+    "alias": [],
+    "command": "agent:optimize:accept",
+    "flagAliases": [],
+    "flagChars": ["i", "o"],
+    "flags": ["api-version", "execution-id", "flags-dir", "json", "target-org"],
+    "plugin": "@salesforce/plugin-agent"
+  },
+  {
+    "alias": [],
+    "command": "agent:optimize:results",
+    "flagAliases": [],
+    "flagChars": ["i", "o"],
+    "flags": ["api-version", "execution-id", "flags-dir", "json", "target-org"],
+    "plugin": "@salesforce/plugin-agent"
+  },
+  {
+    "alias": [],
+    "command": "agent:optimize:run",
+    "flagAliases": [],
+    "flagChars": ["i", "o", "s"],
+    "flags": ["api-version", "authoring-bundle", "flags-dir", "iterations", "json", "spec", "target-org"],
+    "plugin": "@salesforce/plugin-agent"
+  },
+  {
+    "alias": [],
+    "command": "agent:optimize:start",
+    "flagAliases": [],
+    "flagChars": ["b", "c", "o", "w"],
+    "flags": [
+      "api-version",
+      "authoring-bundle",
+      "criteria",
+      "flags-dir",
+      "json",
+      "max-iterations",
+      "target-org",
+      "target-score",
+      "test-cases",
+      "wait"
+    ],
+    "plugin": "@salesforce/plugin-agent"
+  },
+  {
+    "alias": [],
+    "command": "agent:optimize:status",
+    "flagAliases": [],
+    "flagChars": ["i", "o"],
+    "flags": ["api-version", "execution-id", "flags-dir", "json", "target-org"],
+    "plugin": "@salesforce/plugin-agent"
+  },
   {
     "alias": [],
     "command": "agent:preview",
diff --git a/messages/agent.optimize.accept.md b/messages/agent.optimize.accept.md
new file mode 100644
index 00000000..31bca3dc
--- /dev/null
+++ b/messages/agent.optimize.accept.md
@@ -0,0 +1,21 @@
+# summary
+
+Accept and publish the results of a completed HEPO optimization.
+
+# description
+
+Sends a publish signal to the optimization workflow, deploying the best-performing agent configuration as a draft.
+
+# examples
+
+- Accept and publish an optimization result:
+
+  <%= config.bin %> <%= command.id %> --execution-id hepo:wf-abc123:run-xyz --target-org myOrg
+
+# flags.execution-id.summary
+
+Execution ID returned by `sf agent optimize start`.
+
+# error.acceptFailed
+
+Failed to accept optimization: %s
diff --git a/messages/agent.optimize.results.md b/messages/agent.optimize.results.md
new file mode 100644
index 00000000..0ea79e42
--- /dev/null
+++ b/messages/agent.optimize.results.md
@@ -0,0 +1,25 @@
+# summary
+
+Get the final results of a completed HEPO optimization run.
+
+# description
+
+Returns detailed results of a completed optimization, including iteration history, score improvements, and the best agent snapshot.
+
+# examples
+
+- Get results of a completed optimization:
+
+  <%= config.bin %> <%= command.id %> --execution-id hepo:wf-abc123:run-xyz --target-org myOrg
+
+# flags.execution-id.summary
+
+Execution ID returned by `sf agent optimize start`.
+
+# error.resultsFailed
+
+Failed to get optimization results: %s
+
+# error.stillRunning
+
+Optimization %s is still running. Use `sf agent optimize status` to check progress.
diff --git a/messages/agent.optimize.run.md b/messages/agent.optimize.run.md
new file mode 100644
index 00000000..cf7977f9
--- /dev/null
+++ b/messages/agent.optimize.run.md
@@ -0,0 +1,95 @@
+# summary
+
+Optimize an Agentforce agent by iteratively improving its instructions.
+
+# description
+
+Runs an optimization loop that evaluates the agent against test cases, uses an LLM to propose instruction improvements, and keeps changes that improve the score.
+
+Each iteration:
+
+1. Sends test utterances to the agent via preview
+2. Scores responses against expected outputs
+3. Proposes instruction edits via an LLM
+4. Applies edits and re-evaluates
+5. Keeps improvements, rejects regressions
+
+Uses the org's Einstein LLM to propose improvements. Falls back to ANTHROPIC_API_KEY or OPENAI_API_KEY from the environment when Einstein is not available on the org.
+
+# flags.authoring-bundle.summary
+
+Name of the authoring bundle to optimize.
+
+# flags.spec.summary
+
+Path to optimization spec file (JSON or YAML). Contains test cases with utterances and expected response keywords.
+
+# flags.iterations.summary
+
+Number of optimization iterations to run (default: 3).
+
+# examples
+
+- Optimize an agent with 3 iterations:
+
+  <%= config.bin %> <%= command.id %> --authoring-bundle MyAgent --spec test-cases.json --target-org my-org
+
+- Run 5 optimization iterations:
+
+  <%= config.bin %> <%= command.id %> --authoring-bundle MyAgent --spec test-cases.json --target-org my-org --iterations 5
+
+# output.baseline
+
+Baseline score: %s (%s/%s test cases passed)
+
+# output.iterationKeep
+
+Iteration %s: KEEP — score improved %s → %s (%s)
+
+# output.iterationReject
+
+Iteration %s: REJECT — score %s did not improve over %s (%s)
+
+# output.summary
+
+Optimization complete. Baseline: %s → Best: %s (%s iterations, %s kept)
+
+# output.noImprovement
+
+No improvements found after %s iterations. Agent unchanged.
+
+# output.alreadyPerfect
+
+All test cases already passing. Agent is fully optimized — no changes needed.
+
+# output.agentUpdated
+
+Agent file updated: %s
+
+# output.agentRestored
+
+Agent file restored to best version.
+
+# error.specNotFound
+
+Spec file not found: %s.
+
+# error.invalidSpec
+
+Invalid optimization spec: %s. Expected JSON with a "test_cases" array.
+
+# error.bundleNotFound
+
+Authoring bundle '%s' not found in the project.
+
+# error.agentFileNotFound
+
+Agent file not found in bundle '%s'. Expected <BundleName>.agent inside the authoring bundle directory.
+
+# error.llmCallFailed
+
+Einstein LLM call failed: %s.
+
+# error.previewFailed
+
+Preview session failed: %s.
diff --git a/messages/agent.optimize.start.md b/messages/agent.optimize.start.md
new file mode 100644
index 00000000..4dfe3431
--- /dev/null
+++ b/messages/agent.optimize.start.md
@@ -0,0 +1,83 @@
+# summary
+
+Start a server-side HEPO optimization for an Agentforce agent.
+
+# description
+
+Starts an iterative optimization workflow on the server. The workflow evaluates the agent against test cases, proposes instruction improvements via an LLM, and keeps changes that improve the score.
+
+Unlike `sf agent optimize run` (which runs optimization client-side), this command offloads all work to the HEPO service and only polls for progress.
+
+# examples
+
+- Start optimization and wait for completion:
+
+  <%= config.bin %> <%= command.id %> --authoring-bundle ShoppingAgent --criteria criteria.yaml --target-org myOrg --wait 60
+
+- Start optimization without waiting (check status later):
+
+  <%= config.bin %> <%= command.id %> --authoring-bundle ShoppingAgent --criteria criteria.yaml --target-org myOrg
+
+- Start with custom iteration settings:
+
+  <%= config.bin %> <%= command.id %> --authoring-bundle ShoppingAgent --criteria criteria.yaml --max-iterations 15 --target-score 0.95 --target-org myOrg
+
+# flags.authoring-bundle.summary
+
+Name of the agent authoring bundle to optimize.
+
+# flags.criteria.summary
+
+Path to criteria YAML file defining scoring metrics, gates, and weights.
+
+# flags.test-cases.summary
+
+Path to test cases YAML file. If not provided, test cases from the criteria file are used.
+
+# flags.max-iterations.summary
+
+Maximum number of optimization iterations (default: 10).
+
+# flags.target-score.summary
+
+Target composite score to stop early (default: 1.0).
+
+# flags.wait.summary
+
+Minutes to wait for completion.
+
+# flags.wait.description
+
+Poll for status updates until the optimization completes or times out. Without --wait, the command returns immediately after starting.
+
+# output.started
+
+Optimization started. Execution ID: %s
+
+# output.progress
+
+Iteration %s/%s — Baseline: %s Current: %s Best: %s
+
+# output.completed
+
+Optimization completed in %s iterations. Baseline: %s → Best: %s
+
+# output.timeout
+
+Optimization still running after %s minutes. Use `sf agent optimize status --execution-id %s` to check progress.
+
+# error.startFailed
+
+Failed to start optimization: %s
+
+# error.pollFailed
+
+Failed to poll optimization status: %s
+
+# error.criteriaNotFound
+
+Criteria file not found: %s
+
+# error.invalidCriteria
+
+Invalid criteria file: %s
diff --git a/messages/agent.optimize.status.md b/messages/agent.optimize.status.md
new file mode 100644
index 00000000..36d34ca6
--- /dev/null
+++ b/messages/agent.optimize.status.md
@@ -0,0 +1,21 @@
+# summary
+
+Get the status of a server-side HEPO optimization run.
+
+# description
+
+Returns the current status of a running or completed HEPO optimization, including iteration progress and score improvements.
+
+# examples
+
+- Get status of an optimization run:
+
+  <%= config.bin %> <%= command.id %> --execution-id hepo:wf-abc123:run-xyz --target-org myOrg
+
+# flags.execution-id.summary
+
+Execution ID returned by `sf agent optimize start`.
+
+# error.statusFailed
+
+Failed to get optimization status: %s
diff --git a/package.json b/package.json
index e3dcb5fc..d8af35d5 100644
--- a/package.json
+++ b/package.json
@@ -91,6 +91,10 @@
             "description": "Commands to generate agent artifacts, such as the agent spec YAML file, authoring bundle, and test spec file.",
             "external": true
           },
+          "optimize": {
+            "description": "Commands to optimize agents.",
+            "external": true
+          },
           "validate": {
             "description": "Command to validate an Agent Script file.",
             "external": true
@@ -258,5 +262,6 @@
     }
   },
   "exports": "./lib/index.js",
-  "type": "module"
+  "type": "module",
+  "packageManager": "yarn@1.22.22+sha512.a6b2f7906b721bba3d67d4aff083df04dad64c399707841b7acf00f6b133b7ac24255f2652fa22ae3534329dc6180534e98d17432037ff6fd140556e2bb3137e"
 }
diff --git a/schemas/agent-optimize-accept.json b/schemas/agent-optimize-accept.json
new file mode 100644
index 00000000..d97d648a
--- /dev/null
+++ b/schemas/agent-optimize-accept.json
@@ -0,0 +1,25 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$ref": "#/definitions/AgentOptimizeAcceptResult",
+  "definitions": {
+    "AgentOptimizeAcceptResult": {
+      "$ref": "#/definitions/OptimizationAcceptResult"
+    },
+    "OptimizationAcceptResult": {
+      "type": "object",
+      "properties": {
+        "executionId": {
+          "type": "string"
+        },
+        "published": {
+          "type": "boolean"
+        },
+        "message": {
+          "type": "string"
+        }
+      },
+      "required": ["executionId", "published", "message"],
+      "additionalProperties": false
+    }
+  }
+}
diff --git a/schemas/agent-optimize-results.json b/schemas/agent-optimize-results.json
new file mode 100644
index 00000000..11ded123
--- /dev/null
+++ b/schemas/agent-optimize-results.json
@@ -0,0 +1,68 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$ref": "#/definitions/AgentOptimizeResultsResult",
+  "definitions": {
+    "AgentOptimizeResultsResult": {
+      "$ref": "#/definitions/OptimizationResults"
+    },
+    "OptimizationResults": {
+      "type": "object",
+      "properties": {
+        "executionId": {
+          "type": "string"
+        },
+        "status": {
+          "type": "string"
+        },
+        "iterationsRun": {
+          "type": "number"
+        },
+        "baselineScore": {
+          "type": "number"
+        },
+        "finalScore": {
+          "type": "number"
+        },
+        "bestScore": {
+          "type": "number"
+        },
+        "bestIteration": {
+          "type": "number"
+        },
+        "iterationHistory": {
+          "type": "array",
+          "items": {
+            "$ref": "#/definitions/IterationEntry"
+          }
+        }
+      },
+      "required": [
+        "executionId",
+        "status",
+        "iterationsRun",
+        "baselineScore",
+        "finalScore",
+        "bestScore",
+        "bestIteration",
+        "iterationHistory"
+      ],
+      "additionalProperties": false
+    },
+    "IterationEntry": {
+      "type": "object",
+      "properties": {
+        "iteration": {
+          "type": "number"
+        },
+        "compositeScore": {
+          "type": "number"
+        },
+        "gatesPassed": {
+          "type": "boolean"
+        }
+      },
+      "required": ["iteration", "compositeScore", "gatesPassed"],
+      "additionalProperties": false
+    }
+  }
+}
diff --git a/schemas/agent-optimize-run.json b/schemas/agent-optimize-run.json
new file mode 100644
index 00000000..571897ef
--- /dev/null
+++ b/schemas/agent-optimize-run.json
@@ -0,0 +1,51 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$ref": "#/definitions/OptimizeRunResult",
+  "definitions": {
+    "OptimizeRunResult": {
+      "type": "object",
+      "properties": {
+        "baselineScore": {
+          "type": "number"
+        },
+        "bestScore": {
+          "type": "number"
+        },
+        "iterations": {
+          "type": "array",
+          "items": {
+            "type": "object",
+            "properties": {
+              "iteration": {
+                "type": "number"
+              },
+              "mutation": {
+                "type": "string"
+              },
+              "score": {
+                "type": "number"
+              },
+              "passCount": {
+                "type": "number"
+              },
+              "totalCount": {
+                "type": "number"
+              },
+              "decision": {
+                "type": "string",
+                "enum": ["KEEP", "REJECT"]
+              }
+            },
+            "required": ["iteration", "mutation", "score", "passCount", "totalCount", "decision"],
+            "additionalProperties": false
+          }
+        },
+        "agentFile": {
+          "type": "string"
+        }
+      },
+      "required": ["baselineScore", "bestScore", "iterations", "agentFile"],
+      "additionalProperties": false
+    }
+  }
+}
diff --git a/schemas/agent-optimize-start.json b/schemas/agent-optimize-start.json
new file mode 100644
index 00000000..23d833e3
--- /dev/null
+++ b/schemas/agent-optimize-start.json
@@ -0,0 +1,28 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$ref": "#/definitions/AgentOptimizeStartResult",
+  "definitions": {
+    "AgentOptimizeStartResult": {
+      "type": "object",
+      "properties": {
+        "executionId": {
+          "type": "string"
+        },
+        "status": {
+          "type": "string"
+        },
+        "baselineScore": {
+          "type": "number"
+        },
+        "bestScore": {
+          "type": "number"
+        },
+        "iterationsRun": {
+          "type": "number"
+        }
+      },
+      "required": ["executionId", "status"],
+      "additionalProperties": false
+    }
+  }
+}
diff --git a/schemas/agent-optimize-status.json b/schemas/agent-optimize-status.json
new file mode 100644
index 00000000..109776cf
--- /dev/null
+++ b/schemas/agent-optimize-status.json
@@ -0,0 +1,57 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$ref": "#/definitions/AgentOptimizeStatusResult",
+  "definitions": {
+    "AgentOptimizeStatusResult": {
+      "$ref": "#/definitions/OptimizationStatus"
+    },
+    "OptimizationStatus": {
+      "type": "object",
+      "properties": {
+        "executionId": {
+          "type": "string"
+        },
+        "status": {
+          "$ref": "#/definitions/OptimizationStatusEnum"
+        },
+        "currentIteration": {
+          "type": "number"
+        },
+        "maxIterations": {
+          "type": "number"
+        },
+        "baselineScore": {
+          "type": "number"
+        },
+        "currentScore": {
+          "type": "number"
+        },
+        "bestScore": {
+          "type": "number"
+        },
+        "bestIteration": {
+          "type": "number"
+        },
+        "message": {
+          "type": "string"
+        }
+      },
+      "required": [
+        "executionId",
+        "status",
+        "currentIteration",
+        "maxIterations",
+        "baselineScore",
+        "currentScore",
+        "bestScore",
+        "bestIteration",
+        "message"
+      ],
+      "additionalProperties": false
+    },
+    "OptimizationStatusEnum": {
+      "type": "string",
+      "enum": ["RUNNING", "COMPLETED", "FAILED", "STOPPED_EARLY"]
+    }
+  }
+}
diff --git a/src/commands/agent/optimize/accept.ts b/src/commands/agent/optimize/accept.ts
new file mode 100644
index 00000000..042e658d
--- /dev/null
+++ b/src/commands/agent/optimize/accept.ts
@@ -0,0 +1,61 @@
+/*
+ * Copyright 2026, Salesforce, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import { SfCommand, Flags } from '@salesforce/sf-plugins-core';
+import { Messages, SfError } from '@salesforce/core';
+import { AgentOptimization, type OptimizationAcceptResult } from '@salesforce/agents';
+
+Messages.importMessagesDirectoryFromMetaUrl(import.meta.url);
+const messages = Messages.loadMessages('@salesforce/plugin-agent', 'agent.optimize.accept');
+
+export type AgentOptimizeAcceptResult = OptimizationAcceptResult;
+
+export default class AgentOptimizeAccept extends SfCommand<AgentOptimizeAcceptResult> {
+  public static readonly summary = messages.getMessage('summary');
+  public static readonly description = messages.getMessage('description');
+  public static readonly examples = messages.getMessages('examples');
+  public static readonly enableJsonFlag = true;
+
+  public static readonly flags = {
+    'target-org': Flags.requiredOrg(),
+    'api-version': Flags.orgApiVersion(),
+    'execution-id': Flags.string({
+      summary: messages.getMessage('flags.execution-id.summary'),
+      char: 'i',
+      required: true,
+    }),
+  };
+
+  public async run(): Promise<AgentOptimizeAcceptResult> {
+    const { flags } = await this.parse(AgentOptimizeAccept);
+    const connection = flags['target-org'].getConnection(flags['api-version']);
+
+    let result: OptimizationAcceptResult;
+    try {
+      result = await AgentOptimization.accept(connection, flags['execution-id']);
+    } catch (error) {
+      const wrapped = SfError.wrap(error);
+      throw new SfError(messages.getMessage('error.acceptFailed', [wrapped.message]), 'AcceptFailed', [], 4, wrapped);
+    }
+
+    if (result.published) {
+      this.log(`Optimization ${result.executionId} accepted and published.`);
+    } else {
+      this.log(`Optimization ${result.executionId}: ${result.message}`);
+    }
+
+    return result;
+  }
+}
diff --git a/src/commands/agent/optimize/results.ts b/src/commands/agent/optimize/results.ts
new file mode 100644
index 00000000..4de80eb0
--- /dev/null
+++ b/src/commands/agent/optimize/results.ts
@@ -0,0 +1,83 @@
+/*
+ * Copyright 2026, Salesforce, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import { SfCommand, Flags } from '@salesforce/sf-plugins-core';
+import { Messages, SfError } from '@salesforce/core';
+import { AgentOptimization, type OptimizationResults } from '@salesforce/agents';
+
+Messages.importMessagesDirectoryFromMetaUrl(import.meta.url);
+const messages = Messages.loadMessages('@salesforce/plugin-agent', 'agent.optimize.results');
+
+export type AgentOptimizeResultsResult = OptimizationResults;
+
+export default class AgentOptimizeResults extends SfCommand<AgentOptimizeResultsResult> {
+  public static readonly summary = messages.getMessage('summary');
+  public static readonly description = messages.getMessage('description');
+  public static readonly examples = messages.getMessages('examples');
+  public static readonly enableJsonFlag = true;
+
+  public static readonly flags = {
+    'target-org': Flags.requiredOrg(),
+    'api-version': Flags.orgApiVersion(),
+    'execution-id': Flags.string({
+      summary: messages.getMessage('flags.execution-id.summary'),
+      char: 'i',
+      required: true,
+    }),
+  };
+
+  public async run(): Promise<AgentOptimizeResultsResult> {
+    const { flags } = await this.parse(AgentOptimizeResults);
+    const connection = flags['target-org'].getConnection(flags['api-version']);
+
+    let result: OptimizationResults;
+    try {
+      result = await AgentOptimization.results(connection, flags['execution-id']);
+    } catch (error) {
+      const wrapped = SfError.wrap(error);
+      if (wrapped.message.includes('still running')) {
+        throw new SfError(
+          messages.getMessage('error.stillRunning', [flags['execution-id']]),
+          'StillRunning',
+          ['Use `sf agent optimize status` to check progress.'],
+          4,
+          wrapped
+        );
+      }
+      throw new SfError(messages.getMessage('error.resultsFailed', [wrapped.message]), 'ResultsFailed', [], 4, wrapped);
+    }
+
+    this.log(`Execution: ${result.executionId}`);
+    this.log(`Status: ${result.status}`);
+    this.log(`Iterations: ${result.iterationsRun}`);
+    this.log(
+      `Baseline: ${result.baselineScore.toFixed(2)} → Final: ${result.finalScore.toFixed(
+        2
+      )} (Best: ${result.bestScore.toFixed(2)} at iteration ${result.bestIteration})`
+    );
+
+    if (result.iterationHistory?.length) {
+      this.log('\nIteration History:');
+      this.log('  #   Score   Gates');
+      this.log('  --- ------- -----');
+      for (const iter of result.iterationHistory) {
+        const gates = iter.gatesPassed ? 'PASS' : 'FAIL';
+        this.log(`  ${String(iter.iteration).padStart(3)}   ${iter.compositeScore.toFixed(3).padStart(7)}   ${gates}`);
+      }
+    }
+
+    return result;
+  }
+}
diff --git a/src/commands/agent/optimize/run.ts b/src/commands/agent/optimize/run.ts
new file mode 100644
index 00000000..86aa55e6
--- /dev/null
+++ b/src/commands/agent/optimize/run.ts
@@ -0,0 +1,532 @@
+/*
+ * Copyright 2026, Salesforce, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+import { readFile, writeFile } from 'node:fs/promises';
+import { join } from 'node:path';
+import { Flags, SfCommand, toHelpSection } from '@salesforce/sf-plugins-core';
+import { Connection, EnvironmentVariable, Messages, SfError, SfProject } from '@salesforce/core';
+import { Agent, findAuthoringBundle } from '@salesforce/agents';
+import { parse as parseYaml } from 'yaml';
+
+Messages.importMessagesDirectoryFromMetaUrl(import.meta.url);
+const messages = Messages.loadMessages('@salesforce/plugin-agent', 'agent.optimize.run');
+
+// ---------------------------------------------------------------------------
+// Types
+// ---------------------------------------------------------------------------
+
+type TestCase = {
+  utterance: string;
+  expected_response_contains?: string[];
+  expected_topic?: string;
+};
+
+type OptimizationSpec = {
+  test_cases: TestCase[];
+};
+
+type TestCaseResult = {
+  utterance: string;
+  response: string;
+  passed: boolean;
+  matchedKeywords: string[];
+  missedKeywords: string[];
+};
+
+type IterationResult = {
+  iteration: number;
+  mutation: string;
+  score: number;
+  passCount: number;
+  totalCount: number;
+  decision: 'KEEP' | 'REJECT';
+};
+
+export type OptimizeRunResult = {
+  baselineScore: number;
+  bestScore: number;
+  iterations: IterationResult[];
+  agentFile: string;
+};
+
+// ---------------------------------------------------------------------------
+// LLM client — tries Einstein Prompt Generations API first, then falls back
+// to ANTHROPIC_API_KEY or OPENAI_API_KEY from the environment.
+// ---------------------------------------------------------------------------
+
+type PromptGenerationResponse = {
+  generations: Array<{ text: string; responseId: string }>;
+};
+
+type AnthropicResponse = {
+  content: Array<{ type: string; text: string }>;
+};
+
+type OpenAIResponse = {
+  choices: Array<{ message: { content: string } }>;
+};
+
+async function callLlmEinstein(conn: Connection, prompt: string): Promise<string> {
+  const body = {
+    isPreview: false,
+    inputParams: {
+      valueMap: {
+        'Input:prompt_text': {
+          value: prompt,
+        },
+      },
+    },
+    additionalConfig: {
+      applicationName: 'PromptTemplateGenerationsInvocable',
+      maxTokens: 4096,
+      temperature: 0,
+    },
+  };
+
+  const url = `/services/data/v${conn.version}/einstein/prompt-templates/AgentOptimizer/generations`;
+
+  const resp = await conn.request<PromptGenerationResponse>({
+    method: 'POST',
+    url,
+    body: JSON.stringify(body),
+  });
+
+  if (!resp.generations?.length) {
+    throw new Error('Empty response from Einstein LLM');
+  }
+  return resp.generations[0].text;
+}
+
+async function callLlmAnthropic(apiKey: string, prompt: string): Promise<string> {
+  const resp = await fetch('https://api.anthropic.com/v1/messages', {
+    method: 'POST',
+    headers: {
+      'x-api-key': apiKey,
+      'anthropic-version': '2023-06-01',
+      'content-type': 'application/json',
+    },
+    body: JSON.stringify({
+      model: 'claude-sonnet-4-20250514',
+      // eslint-disable-next-line camelcase
+      max_tokens: 4096,
+      messages: [{ role: 'user', content: prompt }],
+    }),
+  });
+  if (!resp.ok) throw new Error(`Anthropic API error: ${resp.status} ${resp.statusText}`);
+  const data = (await resp.json()) as AnthropicResponse;
+  return data.content[0].text;
+}
+
+async function callLlmOpenAI(apiKey: string, prompt: string): Promise<string> {
+  const resp = await fetch('https://api.openai.com/v1/chat/completions', {
+    method: 'POST',
+    headers: {
+      Authorization: `Bearer ${apiKey}`,
+      'Content-Type': 'application/json',
+    },
+    body: JSON.stringify({
+      model: 'gpt-4o',
+      // eslint-disable-next-line camelcase
+      max_tokens: 4096,
+      temperature: 0,
+      messages: [{ role: 'user', content: prompt }],
+    }),
+  });
+  if (!resp.ok) throw new Error(`OpenAI API error: ${resp.status} ${resp.statusText}`);
+  const data = (await resp.json()) as OpenAIResponse;
+  return data.choices[0].message.content;
+}
+
+async function callLlm(conn: Connection, prompt: string): Promise<string> {
+  // Try Einstein first
+  try {
+    return await callLlmEinstein(conn, prompt);
+  } catch {
+    // Einstein not available — fall through to external providers
+  }
+
+  // Fallback: ANTHROPIC_API_KEY
+  const anthropicKey = process.env.ANTHROPIC_API_KEY;
+  if (anthropicKey) {
+    return callLlmAnthropic(anthropicKey, prompt);
+  }
+
+  // Fallback: OPENAI_API_KEY
+  const openaiKey = process.env.OPENAI_API_KEY;
+  if (openaiKey) {
+    return callLlmOpenAI(openaiKey, prompt);
+  }
+
+  throw new SfError(messages.getMessage('error.llmCallFailed', ['No LLM provider available']), 'LlmProviderError', [
+    'Einstein Prompt Generations API is not available on this org.',
+    'Set ANTHROPIC_API_KEY or OPENAI_API_KEY environment variable as a fallback.',
+    'Or enable Einstein generative AI and assign EinsteinGPTPromptTemplateUser permission set.',
+  ]);
+}
+
+// ---------------------------------------------------------------------------
+// Agent file helpers
+// ---------------------------------------------------------------------------
+
+function locateAgentFile(project: SfProject, bundleName: string): string {
+  const dirs = project.getPackageDirectories().map((d) => d.fullPath);
+  const bundleDir = findAuthoringBundle(dirs, bundleName);
+  if (!bundleDir) {
+    throw new SfError(messages.getMessage('error.bundleNotFound', [bundleName]));
+  }
+  return join(bundleDir, `${bundleName}.agent`);
+}
+
+// ---------------------------------------------------------------------------
+// Scoring
+// ---------------------------------------------------------------------------
+
+async function evaluateAgent(
+  conn: Connection,
+  project: SfProject,
+  bundleName: string,
+  testCases: TestCase[],
+  log: (msg: string) => void
+): Promise<{ score: number; passCount: number; results: TestCaseResult[] }> {
+  const results: TestCaseResult[] = [];
+
+  // Init once — Agent.init with aabName compiles the .agent file and returns a ScriptAgent
+  const agent = await Agent.init({
+    connection: conn,
+    project,
+    aabName: bundleName,
+  });
+  agent.preview.setMockMode('Mock');
+
+  for (const tc of testCases) {
+    let responseText = '';
+    try {
+      // Start a fresh session per utterance to avoid conversational state leaking
+      // eslint-disable-next-line no-await-in-loop
+      await agent.preview.start({});
+      // eslint-disable-next-line no-await-in-loop
+      const response = await agent.preview.send(tc.utterance);
+      responseText = response.messages
+        .filter((m) => m.type === 'Inform')
+        .map((m) => m.message ?? '')
+        .join(' ');
+      // eslint-disable-next-line no-await-in-loop
+      await agent.preview.end();
+    } catch (e) {
+      log(`  Preview failed for "${tc.utterance}": ${(e as Error).message}`);
+      try {
+        // eslint-disable-next-line no-await-in-loop
+        await agent.preview.end();
+      } catch {
+        // best-effort cleanup
+      }
+    }
+
+    const expectedKeywords = tc.expected_response_contains ?? [];
+    const lowerResponse = responseText.toLowerCase();
+    const matchedKeywords = expectedKeywords.filter((kw) => lowerResponse.includes(kw.toLowerCase()));
+    const missedKeywords = expectedKeywords.filter((kw) => !lowerResponse.includes(kw.toLowerCase()));
+
+    const passed = expectedKeywords.length === 0 || missedKeywords.length === 0;
+
+    results.push({
+      utterance: tc.utterance,
+      response: responseText,
+      passed,
+      matchedKeywords,
+      missedKeywords,
+    });
+  }
+
+  const passCount = results.filter((r) => r.passed).length;
+  const score = testCases.length > 0 ? passCount / testCases.length : 0;
+  return { score, passCount, results };
+}
+
+// ---------------------------------------------------------------------------
+// Mutation proposal
+// ---------------------------------------------------------------------------
+
+function stripCodeFences(text: string): string {
+  let stripped = text.trim();
+  stripped = stripped.replace(/^```(?:yaml|agent|agentscript)?\s*\n?/i, '');
+  stripped = stripped.replace(/\n?```\s*$/, '');
+  return stripped.trim();
+}
+
+function buildMutationPrompt(agentContent: string, evalResults: TestCaseResult[]): string {
+  const failures = evalResults
+    .filter((r) => !r.passed)
+    .map(
+      (r) =>
+        `- Utterance: "${r.utterance}"\n  Response: "${r.response.slice(
+          0,
+          200
+        )}"\n  Missing keywords: ${r.missedKeywords.join(', ')}`
+    )
+    .join('\n');
+
+  const successes = evalResults
+    .filter((r) => r.passed)
+    .map((r) => `- Utterance: "${r.utterance}" — PASSED`)
+    .join('\n');
+
+  return `You are an Agentforce agent optimization expert. You must improve the agent's instructions to fix failing test cases while preserving passing ones.
+
+CRITICAL RULES:
+- Output ONLY the raw agent file content. No markdown fences, no explanations, no commentary.
+- Preserve the EXACT file structure: system:, config:, variables:, language:, start_agent, topic blocks.
+- Only modify text inside "instructions:" fields and "description:" fields.
+- Do NOT add new topics, variables, or actions unless absolutely necessary.
+- Do NOT change the config: block or default_agent_user.
+- Keep all indentation and block-scalar markers (| and ->) exactly as they are.
+
+## Current Agent File (this is valid AgentScript — preserve its structure)
+${agentContent}
+
+## Test Results
+### Failing (needs improvement):
+${failures || '(none)'}
+
+### Passing (preserve these):
+${successes || '(none)'}
+
+## What to change
+Improve ONLY the reasoning instructions text to better handle failing utterances. For example:
+- Add explicit routing rules like "Route return/refund requests to the returns topic"
+- Add keywords the agent should mention in responses (e.g., "Always mention refund policy when discussing returns")
+- Make topic descriptions more specific so routing works correctly
+
+Output the complete agent file now:`;
+}
+
+// ---------------------------------------------------------------------------
+// Command
+// ---------------------------------------------------------------------------
+
+export default class AgentOptimizeRun extends SfCommand<OptimizeRunResult> {
+  public static readonly summary = messages.getMessage('summary');
+  public static readonly description = messages.getMessage('description');
+  public static readonly examples = messages.getMessages('examples');
+  public static readonly requiresProject = true;
+  public static state = 'beta';
+
+  public static readonly envVariablesSection = toHelpSection(
+    'ENVIRONMENT VARIABLES',
+    EnvironmentVariable.SF_TARGET_ORG
+  );
+
+  public static readonly flags = {
+    'target-org': Flags.requiredOrg(),
+    'api-version': Flags.orgApiVersion(),
+    'authoring-bundle': Flags.string({
+      summary: messages.getMessage('flags.authoring-bundle.summary'),
+      required: true,
+    }),
+    spec: Flags.file({
+      char: 's',
+      required: true,
+      summary: messages.getMessage('flags.spec.summary'),
+      exists: true,
+    }),
+    iterations: Flags.integer({
+      char: 'i',
+      default: 3,
+      summary: messages.getMessage('flags.iterations.summary'),
+    }),
+  };
+
+  public async run(): Promise<OptimizeRunResult> {
+    const { flags } = await this.parse(AgentOptimizeRun);
+    const org = flags['target-org'];
+    const conn = org.getConnection(flags['api-version']);
+    const bundleName = flags['authoring-bundle'];
+    const maxIterations = flags.iterations;
+
+    // 1. Load spec (supports JSON and YAML)
+    let spec: OptimizationSpec;
+    try {
+      const raw = await readFile(flags.spec, 'utf-8');
+      if (flags.spec.endsWith('.yml') || flags.spec.endsWith('.yaml')) {
+        spec = parseYaml(raw) as OptimizationSpec;
+      } else {
+        spec = JSON.parse(raw) as OptimizationSpec;
+      }
+    } catch (e) {
+      throw new SfError(messages.getMessage('error.invalidSpec', [(e as Error).message]));
+    }
+    if (!spec.test_cases || !Array.isArray(spec.test_cases) || spec.test_cases.length === 0) {
+      throw new SfError(messages.getMessage('error.invalidSpec', ['missing or empty "test_cases" array']));
+    }
+
+    // 2. Locate agent file
+    const agentFile = locateAgentFile(this.project!, bundleName);
+    let agentContent: string;
+    try {
+      agentContent = await readFile(agentFile, 'utf-8');
+    } catch {
+      throw new SfError(messages.getMessage('error.agentFileNotFound', [bundleName]));
+    }
+    const originalContent = agentContent;
+
+    // 3. Baseline evaluation
+    this.log(`Evaluating baseline (${spec.test_cases.length} test cases)...`);
+    const baseline = await evaluateAgent(conn, this.project!, bundleName, spec.test_cases, (msg) => this.log(msg));
+    this.log(
+      messages.getMessage('output.baseline', [
+        `${(baseline.score * 100).toFixed(0)}%`,
+        baseline.passCount.toString(),
+        spec.test_cases.length.toString(),
+      ])
+    );
+
+    // 4. Optimization loop
+    let bestScore = baseline.score;
+    let bestContent = agentContent;
+    let currentResults = baseline.results;
+    const iterations: IterationResult[] = [];
+    let keptCount = 0;
+
+    for (let i = 1; i <= maxIterations; i++) {
+      this.log(`\nIteration ${i}/${maxIterations}...`);
+
+      // 4a. If all tests pass, stop early
+      if (bestScore >= 1.0) {
+        this.log('All test cases passing — stopping early.');
+        break;
+      }
+
+      // 4b. Ask LLM for mutation
+      this.log('  Proposing improvements via LLM...');
+      let mutatedContent: string;
+      try {
+        const prompt = buildMutationPrompt(agentContent, currentResults);
+        // eslint-disable-next-line no-await-in-loop
+        mutatedContent = await callLlm(conn, prompt);
+        mutatedContent = stripCodeFences(mutatedContent);
+      } catch (e) {
+        this.log(`  LLM call failed: ${(e as Error).message}. Skipping iteration.`);
+        iterations.push({
+          iteration: i,
+          mutation: 'LLM_ERROR',
+          score: bestScore,
+          passCount: baseline.passCount,
+          totalCount: spec.test_cases.length,
+          decision: 'REJECT',
+        });
+        continue;
+      }
+
+      // 4c. Basic validation: ensure it looks like an agent file
+      if (!mutatedContent.includes('system:') || !mutatedContent.includes('start_agent')) {
+        this.log('  LLM returned invalid agent content (missing system: or start_agent). Skipping iteration.');
+        iterations.push({
+          iteration: i,
+          mutation: 'INVALID_CONTENT',
+          score: bestScore,
+          passCount: baseline.passCount,
+          totalCount: spec.test_cases.length,
+          decision: 'REJECT',
+        });
+        continue;
+      }
+
+      // 4d. Apply mutation
+      // eslint-disable-next-line no-await-in-loop
+      await writeFile(agentFile, mutatedContent, 'utf-8');
+
+      // 4e. Evaluate
+      this.log('  Evaluating...');
+      // eslint-disable-next-line no-await-in-loop
+      const evalResult = await evaluateAgent(conn, this.project!, bundleName, spec.test_cases, (msg) => this.log(msg));
+
+      // 4f. KEEP or REJECT
+      const mutationLabel = 'instruction_refine';
+      if (evalResult.score > bestScore) {
+        bestScore = evalResult.score;
+        bestContent = mutatedContent;
+        agentContent = mutatedContent;
+        currentResults = evalResult.results;
+        keptCount++;
+        this.log(
+          messages.getMessage('output.iterationKeep', [
+            i.toString(),
+            `${(evalResult.score * 100).toFixed(0)}%`,
+            `${(bestScore * 100).toFixed(0)}%`,
+            mutationLabel,
+          ])
+        );
+        iterations.push({
+          iteration: i,
+          mutation: mutationLabel,
+          score: evalResult.score,
+          passCount: evalResult.passCount,
+          totalCount: spec.test_cases.length,
+          decision: 'KEEP',
+        });
+      } else {
+        // Revert to best known version
+        // eslint-disable-next-line no-await-in-loop
+        await writeFile(agentFile, bestContent, 'utf-8');
+        agentContent = bestContent;
+        this.log(
+          messages.getMessage('output.iterationReject', [
+            i.toString(),
+            `${(evalResult.score * 100).toFixed(0)}%`,
+            `${(bestScore * 100).toFixed(0)}%`,
+            mutationLabel,
+          ])
+        );
+        iterations.push({
+          iteration: i,
+          mutation: mutationLabel,
+          score: evalResult.score,
+          passCount: evalResult.passCount,
+          totalCount: spec.test_cases.length,
+          decision: 'REJECT',
+        });
+      }
+    }
+
+    // 5. Final write of best content
+    await writeFile(agentFile, bestContent, 'utf-8');
+
+    // 6. Summary
+    if (baseline.score >= 1.0 && iterations.length === 0) {
+      this.log(messages.getMessage('output.alreadyPerfect'));
+    } else if (bestScore > baseline.score) {
+      this.log(
+        `\n${messages.getMessage('output.summary', [
+          `${(baseline.score * 100).toFixed(0)}%`,
+          `${(bestScore * 100).toFixed(0)}%`,
+          iterations.length.toString(),
+          keptCount.toString(),
+        ])}`
+      );
+      this.log(messages.getMessage('output.agentUpdated', [agentFile]));
+    } else {
+      await writeFile(agentFile, originalContent, 'utf-8');
+      this.log(`\n${messages.getMessage('output.noImprovement', [iterations.length.toString()])}`);
+    }
+
+    return {
+      baselineScore: baseline.score,
+      bestScore,
+      iterations,
+      agentFile,
+    };
+  }
+}
diff --git a/src/commands/agent/optimize/start.ts b/src/commands/agent/optimize/start.ts
new file mode 100644
index 00000000..c4337c0a
--- /dev/null
+++ b/src/commands/agent/optimize/start.ts
@@ -0,0 +1,167 @@
+/*
+ * Copyright 2026, Salesforce, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import { readFile } from 'node:fs/promises';
+import { SfCommand, Flags } from '@salesforce/sf-plugins-core';
+import { Messages, SfError } from '@salesforce/core';
+import { AgentOptimization, type OptimizationExecution, type OptimizationStatus } from '@salesforce/agents';
+
+Messages.importMessagesDirectoryFromMetaUrl(import.meta.url);
+const messages = Messages.loadMessages('@salesforce/plugin-agent', 'agent.optimize.start');
+
+export type AgentOptimizeStartResult = {
+  executionId: string;
+  status: string;
+  baselineScore?: number;
+  bestScore?: number;
+  iterationsRun?: number;
+};
+
+export default class AgentOptimizeStart extends SfCommand<AgentOptimizeStartResult> {
+  public static readonly summary = messages.getMessage('summary');
+  public static readonly description = messages.getMessage('description');
+  public static readonly examples = messages.getMessages('examples');
+  public static readonly enableJsonFlag = true;
+  public static state = 'beta';
+
+  public static readonly flags = {
+    'target-org': Flags.requiredOrg(),
+    'api-version': Flags.orgApiVersion(),
+    'authoring-bundle': Flags.string({
+      summary: messages.getMessage('flags.authoring-bundle.summary'),
+      char: 'b',
+      required: true,
+    }),
+    criteria: Flags.file({
+      summary: messages.getMessage('flags.criteria.summary'),
+      char: 'c',
+      required: true,
+      exists: true,
+    }),
+    'test-cases': Flags.file({
+      summary: messages.getMessage('flags.test-cases.summary'),
+      exists: true,
+    }),
+    'max-iterations': Flags.integer({
+      summary: messages.getMessage('flags.max-iterations.summary'),
+      default: 10,
+    }),
+    'target-score': Flags.string({
+      summary: messages.getMessage('flags.target-score.summary'),
+      default: '1.0',
+    }),
+    wait: Flags.integer({
+      summary: messages.getMessage('flags.wait.summary'),
+      description: messages.getMessage('flags.wait.description'),
+      char: 'w',
+    }),
+  };
+
+  public async run(): Promise<AgentOptimizeStartResult> {
+    const { flags } = await this.parse(AgentOptimizeStart);
+    const connection = flags['target-org'].getConnection(flags['api-version']);
+
+    let criteriaJson: string;
+    try {
+      criteriaJson = await readFile(flags.criteria, 'utf-8');
+    } catch (e) {
+      throw new SfError(messages.getMessage('error.criteriaNotFound', [(e as Error).message]));
+    }
+
+    let testCasesJson: string | undefined;
+    if (flags['test-cases']) {
+      try {
+        testCasesJson = await readFile(flags['test-cases'], 'utf-8');
+      } catch (e) {
+        throw new SfError(messages.getMessage('error.invalidCriteria', [(e as Error).message]));
+      }
+    }
+
+    let execution: OptimizationExecution;
+    try {
+      execution = await AgentOptimization.start(connection, {
+        authoringBundleName: flags['authoring-bundle'],
+        criteriaJson,
+        testCasesJson,
+        maxIterations: flags['max-iterations'],
+        targetScore: parseFloat(flags['target-score']),
+      });
+    } catch (error) {
+      const wrapped = SfError.wrap(error);
+      throw new SfError(messages.getMessage('error.startFailed', [wrapped.message]), 'StartFailed', [], 4, wrapped);
+    }
+
+    this.log(messages.getMessage('output.started', [execution.executionId]));
+
+    if (!flags.wait) {
+      return { executionId: execution.executionId, status: 'NEW' };
+    }
+
+    const pollIntervalMs = 10_000;
+    const deadlineMs = Date.now() + flags.wait * 60_000;
+
+    while (Date.now() < deadlineMs) {
+      // eslint-disable-next-line no-await-in-loop
+      await sleep(pollIntervalMs);
+
+      let status: OptimizationStatus;
+      try {
+        // eslint-disable-next-line no-await-in-loop
+        status = await AgentOptimization.status(connection, execution.executionId);
+      } catch (error) {
+        const wrapped = SfError.wrap(error);
+        throw new SfError(messages.getMessage('error.pollFailed', [wrapped.message]), 'PollFailed', [], 4, wrapped);
+      }
+
+      if (status.currentIteration != null && status.maxIterations != null) {
+        this.log(
+          messages.getMessage('output.progress', [
+            String(status.currentIteration),
+            String(status.maxIterations),
+            status.baselineScore?.toFixed(2) ?? '-',
+            status.currentScore?.toFixed(2) ?? '-',
+            status.bestScore?.toFixed(2) ?? '-',
+          ])
+        );
+      }
+
+      if (status.status === 'COMPLETED' || status.status === 'FAILED' || status.status === 'STOPPED_EARLY') {
+        this.log(
+          messages.getMessage('output.completed', [
+            String(status.currentIteration ?? 0),
+            status.baselineScore?.toFixed(2) ?? '-',
+            status.bestScore?.toFixed(2) ?? '-',
+          ])
+        );
+        return {
+          executionId: execution.executionId,
+          status: status.status,
+          baselineScore: status.baselineScore,
+          bestScore: status.bestScore,
+          iterationsRun: status.currentIteration,
+        };
+      }
+    }
+
+    this.log(messages.getMessage('output.timeout', [String(flags.wait), execution.executionId]));
+    return { executionId: execution.executionId, status: 'TIMEOUT' };
+  }
+}
+
+function sleep(ms: number): Promise<void> {
+  return new Promise((resolve) => {
+    setTimeout(resolve, ms);
+  });
+}
diff --git a/src/commands/agent/optimize/status.ts b/src/commands/agent/optimize/status.ts
new file mode 100644
index 00000000..5914e05e
--- /dev/null
+++ b/src/commands/agent/optimize/status.ts
@@ -0,0 +1,71 @@
+/*
+ * Copyright 2026, Salesforce, Inc.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+import { SfCommand, Flags } from '@salesforce/sf-plugins-core';
+import { Messages, SfError } from '@salesforce/core';
+import { AgentOptimization, type OptimizationStatus } from '@salesforce/agents';
+
+Messages.importMessagesDirectoryFromMetaUrl(import.meta.url);
+const messages = Messages.loadMessages('@salesforce/plugin-agent', 'agent.optimize.status');
+
+export type AgentOptimizeStatusResult = OptimizationStatus;
+
+export default class AgentOptimizeStatus extends SfCommand<AgentOptimizeStatusResult> {
+  public static readonly summary = messages.getMessage('summary');
+  public static readonly description = messages.getMessage('description');
+  public static readonly examples = messages.getMessages('examples');
+  public static readonly enableJsonFlag = true;
+
+  public static readonly flags = {
+    'target-org': Flags.requiredOrg(),
+    'api-version': Flags.orgApiVersion(),
+    'execution-id': Flags.string({
+      summary: messages.getMessage('flags.execution-id.summary'),
+      char: 'i',
+      required: true,
+    }),
+  };
+
+  public async run(): Promise<AgentOptimizeStatusResult> {
+    const { flags } = await this.parse(AgentOptimizeStatus);
+    const connection = flags['target-org'].getConnection(flags['api-version']);
+
+    let result: OptimizationStatus;
+    try {
+      result = await AgentOptimization.status(connection, flags['execution-id']);
+    } catch (error) {
+      const wrapped = SfError.wrap(error);
+      throw new SfError(messages.getMessage('error.statusFailed', [wrapped.message]), 'StatusFailed', [], 4, wrapped);
+    }
+
+    this.log(`Execution: ${result.executionId}`);
+    this.log(`Status: ${result.status}`);
+    if (result.currentIteration != null && result.maxIterations != null) {
+      this.log(`Progress: ${result.currentIteration}/${result.maxIterations} iterations`);
+    }
+    const parts: string[] = [];
+    if (result.baselineScore != null) parts.push(`Baseline: ${result.baselineScore.toFixed(2)}`);
+    if (result.currentScore != null) parts.push(`Current: ${result.currentScore.toFixed(2)}`);
+    if (result.bestScore != null) {
+      let best = `Best: ${result.bestScore.toFixed(2)}`;
+      if (result.bestIteration != null) best += ` (iteration ${result.bestIteration})`;
+      parts.push(best);
+    }
+    if (parts.length > 0) this.log(parts.join('  '));
+    if (result.message) this.log(result.message);
+
+    return result;
+  }
+}
diff --git a/test/mock-projects/agent-generate-template/specs/optimize-spec.json b/test/mock-projects/agent-generate-template/specs/optimize-spec.json
new file mode 100644
index 00000000..62d8ac2e
--- /dev/null
+++ b/test/mock-projects/agent-generate-template/specs/optimize-spec.json
@@ -0,0 +1,20 @@
+{
+  "test_cases": [
+    {
+      "utterance": "What are the resort's check-in hours?",
+      "expected_response_contains": ["check-in", "hours"]
+    },
+    {
+      "utterance": "I need help with my employee schedule",
+      "expected_response_contains": ["schedule"]
+    },
+    {
+      "utterance": "I have a complaint about my room",
+      "expected_response_contains": ["sorry", "help"]
+    },
+    {
+      "utterance": "What is the weather like today?",
+      "expected_response_contains": ["weather"]
+    }
+  ]
+}
diff --git a/test/mock-projects/agent-generate-template/specs/optimize-spec.yml b/test/mock-projects/agent-generate-template/specs/optimize-spec.yml
new file mode 100644
index 00000000..0c3fb0d8
--- /dev/null
+++ b/test/mock-projects/agent-generate-template/specs/optimize-spec.yml
@@ -0,0 +1,18 @@
+test_cases:
+  - utterance: "What are the resort's check-in hours?"
+    expected_response_contains:
+      - check-in
+      - hours
+
+  - utterance: 'I need help with my employee schedule'
+    expected_response_contains:
+      - schedule
+
+  - utterance: 'I have a complaint about my room'
+    expected_response_contains:
+      - sorry
+      - help
+
+  - utterance: 'What is the weather like today?'
+    expected_response_contains:
+      - weather