Fix directive evaluation

1 files changed, 30 insertions, 13 deletions

diff --git a/makima/src/orchestration/directive.rs b/makima/src/orchestration/directive.rs
index 044fce6..9a65035 100644
--- a/makima/src/orchestration/directive.rs
+++ b/makima/src/orchestration/directive.rs
@@ -293,6 +293,17 @@ pub async fn on_contract_completed(
                 .map_err(|e| format!("Failed to get step by contract: {}", e))?;
 
             if let Some(step) = step {
+                // Idempotency: only dispatch monitoring if step is still "running"
+                // (on_step_completed may also fire via the task path)
+                if step.status != "running" {
+                    tracing::info!(
+                        step_id = %step.id,
+                        status = %step.status,
+                        "Skipping step contract completion: step no longer running"
+                    );
+                    return Ok(());
+                }
+
                 let directive = repository::get_directive(pool, directive_id)
                     .await
                     .map_err(|e| format!("Failed to get directive: {}", e))?
@@ -634,6 +645,17 @@ async fn on_step_completed(
         return Ok(());
     };
 
+    // Idempotency: only process if step is still "running"
+    // (on_contract_completed may also fire via the contract path)
+    if step.status != "running" {
+        tracing::info!(
+            step_id = %step.id,
+            status = %step.status,
+            "Skipping on_step_completed: step no longer running"
+        );
+        return Ok(());
+    }
+
     // Get the directive for threshold info
     let directive = repository::get_directive(pool, directive_id)
         .await
@@ -1048,7 +1070,8 @@ async fn dispatch_monitoring(
     // Build evaluation prompt
     let prompt = build_monitoring_prompt(directive, step, step_contract);
 
-    // Create supervisor task
+    // Create monitoring task (NOT a supervisor — regular task that exits when done,
+    // which triggers on_task_completed → on_monitoring_completed automatically)
     let supervisor_task = repository::create_task_for_owner(
         pool,
         owner_id,
@@ -1058,7 +1081,7 @@ async fn dispatch_monitoring(
             description: Some("Evaluate step output against directive criteria".to_string()),
             plan: prompt,
             parent_task_id: None,
-            is_supervisor: true,
+            is_supervisor: false,
             priority: 8,
             repository_url: directive.repository_url.clone(),
             base_branch: directive.base_branch.clone(),
@@ -1075,9 +1098,9 @@ async fn dispatch_monitoring(
         },
     )
     .await
-    .map_err(|e| format!("Failed to create monitoring supervisor task: {}", e))?;
+    .map_err(|e| format!("Failed to create monitoring task: {}", e))?;
 
-    // Link supervisor to contract
+    // Link monitoring task to contract
     repository::update_contract_for_owner(
         pool,
         contract.id,
@@ -1090,9 +1113,9 @@ async fn dispatch_monitoring(
     .await
     .map_err(|e| match e {
         crate::db::repository::RepositoryError::Database(e) => {
-            format!("Failed to link supervisor to monitoring contract: {}", e)
+            format!("Failed to link task to monitoring contract: {}", e)
         }
-        other => format!("Failed to link supervisor to monitoring contract: {:?}", other),
+        other => format!("Failed to link task to monitoring contract: {:?}", other),
     })?;
 
     // Link step to monitoring contract/task
@@ -1191,8 +1214,7 @@ Scoring guidelines:
 - Score < {threshold_yellow}: confidenceLevel = "red", passed = false
 - Be specific in reworkInstructions if the step fails — the step will be re-executed with these instructions.
 
-After writing the evaluation file, mark the contract as complete:
-  makima supervisor complete"#,
+After writing the evaluation file, you are done. Simply exit."#,
         title = directive.title,
         goal = directive.goal,
         requirements = serde_json::to_string_pretty(&directive.requirements).unwrap_or_default(),
@@ -1216,11 +1238,6 @@ async fn on_monitoring_completed(
     task: &Task,
     owner_id: Uuid,
 ) -> Result<(), String> {
-    // Only process supervisor task completions
-    if !task.is_supervisor {
-        return Ok(());
-    }
-
     let Some(directive_id) = contract.directive_id else {
         return Ok(());
     };