feat: harden Python workflow and structured planning

Replace the uv-specific skill with a portable Python skill and rewrite the default user system prompt as a concise, structured Markdown execution policy.

Normalize a single fenced structured-JSON response, retry invalid skill selection with think disabled, validate selected skill output, and make coding plans keep editing and verification in one task.

Document the file_edit patch payload, expose that contract in the tool catalog, and convert malformed patch input into a retryable tool failure instead of terminating the agent task.

Update configuration and runtime tests for the Python skill, exact tool catalog, structured-output retry, and implementation-and-verification planning guidance.

Verified with 82 Agent.Core tests and an agent-mcp Python scheduling scenario with 49 passing unittest cases.

1 files changed, 39 insertions, 22 deletions

diff --git a/src/Agent.Core/Runtime/AgentRuntime.cs b/src/Agent.Core/Runtime/AgentRuntime.cs
index 4aaf646..31a4414 100644
--- a/src/Agent.Core/Runtime/AgentRuntime.cs
+++ b/src/Agent.Core/Runtime/AgentRuntime.cs
@@ -426,34 +426,46 @@ public sealed class AgentRuntime(TaskRegistry registry, IOllamaClient ollama, Ag
         string catalog = string.Join("\n", settings.Skills.Select(static s => $"- {s.Name}: {s.Description}"));
         var selectionMessages = new List<OllamaMessage>
         {
-            new("system", $"Select skills relevant to the user request. Available skills:\n{catalog}\n\nReturn a JSON object with a 'selected_skills' array containing skill names. Select only relevant skills. If none are relevant, return an empty array."),
-            new("user", task.Request)
+            new("system", $"You are a skill classifier, not a task executor. Do not solve the request, inspect files, write code, or call Tools. Available skills:\n{catalog}\n\nReturn only one JSON object with a 'selected_skills' array containing exact skill names. Select only relevant skills. If none are relevant, return an empty array."),
+            new("user", $"Classify skills for this request only:\n{task.Request}")
         };
 
         string schema = """{"type":"object","properties":{"selected_skills":{"type":"array","items":{"type":"string"}}},"required":["selected_skills"]}""";
         using JsonDocument formatDoc = JsonDocument.Parse(schema);
 
-        OllamaResult result;
-        AgentLlmCallContext? callContext = null;
-        try
+        IReadOnlyList<OllamaMessage> attemptMessages = selectionMessages;
+        for (int attempt = 0; attempt < 2; attempt++)
         {
             if (counters.Iterations >= settings.MaxIterations) return [];
             int iteration = counters.Iterations++;
-            callContext = new AgentLlmCallContext(Guid.NewGuid(), task.SessionId, task.TaskId, null,
+            var callContext = new AgentLlmCallContext(Guid.NewGuid(), task.SessionId, task.TaskId, null,
                 "skill_selection", iteration, counters.Replans, settings.Model);
-            _observer.LlmRequest(callContext, selectionMessages);
-            result = await ollama.ChatStructuredAsync(selectionMessages, task.Temperature, formatDoc.RootElement.Clone(), cancellationToken).ConfigureAwait(false);
-            _observer.LlmResponse(callContext, result);
-            registry.AddUsage(task.TaskId, result.InputTokens, result.OutputTokens);
-            registry.UpdateDiagnostics(task.TaskId, counters.Iterations, counters.Replans);
-        }
-        catch (Exception ex) when (ex is not (OutOfMemoryException or StackOverflowException or AccessViolationException))
-        {
-            if (callContext is not null) _observer.LlmFailure(callContext, "skill_selection_failed", ex.Message);
-            return [];
+            _observer.LlmRequest(callContext, attemptMessages);
+            OllamaResult? result = null;
+            try
+            {
+                bool? thinkOverride = attempt == 0 ? null : false;
+                result = await ollama.ChatStructuredAsync(attemptMessages, task.Temperature, formatDoc.RootElement.Clone(), thinkOverride, cancellationToken).ConfigureAwait(false);
+                _observer.LlmResponse(callContext, result);
+                registry.AddUsage(task.TaskId, result.InputTokens, result.OutputTokens);
+                registry.UpdateDiagnostics(task.TaskId, counters.Iterations, counters.Replans);
+                if (TryParseSelectedSkills(result.Content, out IReadOnlyList<AgentSkill> selected)) return selected;
+                throw new PlannerValidationException("Ollama returned an invalid skill selection object.");
+            }
+            catch (OperationCanceledException) when (cancellationToken.IsCancellationRequested) { throw; }
+            catch (Exception ex) when (ex is not (OutOfMemoryException or StackOverflowException or AccessViolationException))
+            {
+                _observer.LlmFailure(callContext, "skill_selection_failed", ex.Message);
+                if (attempt == 0)
+                {
+                    attemptMessages = selectionMessages.Concat([
+                        new OllamaMessage("system", "The previous response was invalid. Return only {\"selected_skills\":[\"name\"]} or {\"selected_skills\":[]}, with no Markdown, code, Tool call, or commentary.")
+                    ]).ToArray();
+                }
+            }
         }
 
-        return ParseSelectedSkills(result.Content);
+        return [];
     }
 
     private void RecordLlmAttempt(TaskSnapshot task, ExecutionCounters counters, string phase,
@@ -525,15 +537,19 @@ public sealed class AgentRuntime(TaskRegistry registry, IOllamaClient ollama, Ag
         });
     }
 
-    private IReadOnlyList<AgentSkill> ParseSelectedSkills(string content)
+    private bool TryParseSelectedSkills(string content, out IReadOnlyList<AgentSkill> selected)
     {
+        selected = [];
         try
         {
-            using JsonDocument doc = JsonDocument.Parse(content);
-            if (!doc.RootElement.TryGetProperty("selected_skills", out JsonElement skillsElement)) return [];
+            using JsonDocument doc = JsonDocument.Parse(StructuredJson.Unwrap(content));
+            if (doc.RootElement.ValueKind != JsonValueKind.Object ||
+                !doc.RootElement.TryGetProperty("selected_skills", out JsonElement skillsElement) ||
+                skillsElement.ValueKind != JsonValueKind.Array) return false;
             var selectedNames = new HashSet<string>(StringComparer.Ordinal);
             foreach (JsonElement item in skillsElement.EnumerateArray())
             {
+                if (item.ValueKind != JsonValueKind.String) return false;
                 string? name = item.GetString();
                 if (!string.IsNullOrWhiteSpace(name)) selectedNames.Add(name);
             }
@@ -546,11 +562,12 @@ public sealed class AgentRuntime(TaskRegistry registry, IOllamaClient ollama, Ag
                     if (valid.Count >= 10) break;
                 }
             }
-            return valid;
+            selected = valid;
+            return true;
         }
         catch
         {
-            return [];
+            return false;
         }
     }