Bug/stale project id parsing

slate.json

@@ -0,0 +1,8 @@
+{
+  "$schema": "https://randomlabs.ai/config.json",
+  "permission": {
+    "*": "allow",
+    "bash": "ask",
+    "edit": "ask"
+  }
+}

src/agent/mod.rs

@@ -690,8 +690,11 @@ pub async fn run_interactive(
                 Some(&current_input),
                 session.plan_mode,
             );
-            let is_generation = prompts::is_generation_query(&current_input);
             let is_planning = session.plan_mode.is_planning();
+            // Inherit generation mode for short follow-up messages ("sure", "yes", "go ahead",
+            // etc.) so the write/shell tool set is not lost between turns.
+            let is_generation = prompts::is_generation_query(&current_input)
+                || (!is_planning && session.last_was_generation && current_input.trim().len() < 60);
 
             // Note: using raw_chat_history directly which preserves Reasoning blocks
             // This is needed for extended thinking to work with multi-turn conversations
@@ -1138,6 +1141,10 @@ pub async fn run_interactive(
                     // Add to conversation history with tool call records
                     conversation_history.add_turn(input.clone(), text.clone(), tool_calls.clone());
 
+                    // Remember whether this turn had generation tools active so short follow-up
+                    // messages ("sure", "go ahead", etc.) don't lose write/shell access.
+                    session.last_was_generation = is_generation;
+
                     // Check if this heavy turn requires immediate compaction
                     // This helps prevent context overflow in subsequent requests
                     if conversation_history.needs_compaction() {

src/agent/prompts/mod.rs

@@ -705,6 +705,18 @@ pub fn is_generation_query(query: &str) -> bool {
         "new feature",
         "develop",
         "code",
+        // Common modification verbs (previously missing)
+        "fix",
+        "update",
+        "add",
+        "change",
+        "modify",
+        "edit",
+        "configure",
+        "setup",
+        "set up",
+        "patch",
+        "install",
         // Plan execution keywords - needed for plan continuation
         "plan",
         "continue",

src/agent/session/mod.rs

@@ -34,6 +34,10 @@ pub struct ChatSession {
     pub token_usage: TokenUsage,
     /// Current planning mode state
     pub plan_mode: PlanMode,
+    /// Whether the previous turn used generation mode (write/shell tools active).
+    /// Used so short follow-up messages ("sure", "go ahead", "yes") inherit the
+    /// tool set from the previous turn instead of losing write/shell access.
+    pub last_was_generation: bool,
     /// Session loaded via /resume command, to be processed by main loop
     pub pending_resume: Option<crate::agent::persistence::ConversationRecord>,
     /// Platform session state (selected project/org context)
@@ -58,6 +62,7 @@ impl ChatSession {
             history: Vec::new(),
             token_usage: TokenUsage::new(),
             plan_mode: PlanMode::default(),
+            last_was_generation: false,
             pending_resume: None,
             platform_session,
         }

src/agent/tools/platform/create_deployment_config.rs

@@ -13,13 +13,12 @@ use crate::platform::api::types::{
     build_cloud_runner_config_v2,
 };
 use crate::platform::api::{PlatformApiClient, PlatformApiError};
+use crate::platform::session::PlatformSession;
 use std::str::FromStr;
 
 /// Arguments for the create deployment config tool
 #[derive(Debug, Deserialize)]
 pub struct CreateDeploymentConfigArgs {
-    /// The project UUID
-    pub project_id: String,
     /// Service name for the deployment
     pub service_name: String,
     /// Repository ID from GitHub integration
@@ -102,7 +101,6 @@ A deployment config defines how to build and deploy a service, including:
 - Auto-deploy settings
 
 **Required Parameters:**
-- project_id: The project UUID
 - service_name: Name for the service (lowercase, hyphens allowed)
 - repository_id: GitHub repository ID (from platform GitHub integration)
 - repository_full_name: Full repo name like "owner/repo"
@@ -138,10 +136,6 @@ A deployment config defines how to build and deploy a service, including:
             parameters: json!({
                 "type": "object",
                 "properties": {
-                    "project_id": {
-                        "type": "string",
-                        "description": "The UUID of the project"
-                    },
                     "service_name": {
                         "type": "string",
                         "description": "Name for the service (lowercase, hyphens allowed)"
@@ -218,27 +212,38 @@ A deployment config defines how to build and deploy a service, including:
                     }
                 },
                 "required": [
-                    "project_id", "service_name", "repository_id", "repository_full_name",
+                    "service_name", "repository_id", "repository_full_name",
                     "port", "branch", "target_type", "provider", "environment_id"
                 ]
             }),
         }
     }
 
     async fn call(&self, args: Self::Args) -> Result<Self::Output, Self::Error> {
-        // Validate required fields
-        if args.project_id.trim().is_empty() {
+        // Load project_id from session (authoritative source — prevents stale IDs from LLM context)
+        let session = match PlatformSession::load() {
+            Ok(s) => s,
+            Err(_) => {
+                return Ok(format_error_for_llm(
+                    "create_deployment_config",
+                    ErrorCategory::InternalError,
+                    "Failed to load platform session",
+                    Some(vec!["Try authenticating with `sync-ctl auth login`"]),
+                ));
+            }
+        };
+
+        if !session.is_project_selected() {
             return Ok(format_error_for_llm(
                 "create_deployment_config",
                 ErrorCategory::ValidationFailed,
-                "project_id cannot be empty",
-                Some(vec![
-                    "Use list_projects to find valid project IDs",
-                    "Use current_context to get the selected project",
-                ]),
+                "No project selected",
+                Some(vec!["Use select_project to choose a project first"]),
             ));
         }
 
+        let project_id = session.project_id.clone().unwrap_or_default();
+
         if args.service_name.trim().is_empty() {
             return Ok(format_error_for_llm(
                 "create_deployment_config",
@@ -316,7 +321,7 @@ A deployment config defines how to build and deploy a service, including:
             if let Some(ref provider) = provider_enum {
                 if matches!(provider, CloudProvider::Gcp | CloudProvider::Azure) {
                     if let Ok(credential) = client
-                        .check_provider_connection(provider, &args.project_id)
+                        .check_provider_connection(provider, &project_id)
                         .await
                     {
                         if let Some(cred) = credential {
@@ -351,7 +356,7 @@ A deployment config defines how to build and deploy a service, including:
         // Note: Send both field name variants (dockerfile/dockerfilePath, context/buildContext)
         // for backend compatibility - different endpoints may expect different field names
         let request = CreateDeploymentConfigRequest {
-            project_id: args.project_id.clone(),
+            project_id,
             service_name: args.service_name.clone(),
             repository_id: args.repository_id,
             repository_full_name: args.repository_full_name.clone(),