feat: step artifacts framework

- Add Artifact type to Step (name, path, artifact_type, description)
- step_done tool accepts optional artifacts parameter
- Save artifacts to step_artifacts DB table
- Display artifacts in frontend PlanSection (tag style)
- Show artifacts in step context for sub-agents and coordinator
- Add LLM client retry with exponential backoff

src/agent.rs

@@ -12,7 +12,7 @@ use crate::template::{self, LoadedTemplate};
 use crate::tools::ExternalToolManager;
 use crate::LlmConfig;
 
-use crate::state::{AgentState, AgentPhase, Step, StepStatus, StepResult, StepResultStatus, check_scratchpad_size};
+use crate::state::{AgentState, AgentPhase, Artifact, Step, StepStatus, StepResult, StepResultStatus, check_scratchpad_size};
 
 pub struct ServiceInfo {
     pub port: u16,
@@ -47,6 +47,8 @@ pub struct PlanStepInfo {
     pub command: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub status: Option<String>,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub artifacts: Vec<Artifact>,
 }
 
 pub fn plan_infos_from_state(state: &AgentState) -> Vec<PlanStepInfo> {
@@ -63,6 +65,7 @@ pub fn plan_infos_from_state(state: &AgentState) -> Vec<PlanStepInfo> {
             description: s.title.clone(),
             command: s.description.clone(),
             status: Some(status.to_string()),
+            artifacts: s.artifacts.clone(),
         }
     }).collect()
 }
@@ -221,6 +224,15 @@ async fn agent_loop(
                 } else {
                     template::select_template(&llm, &requirement, mgr.template_repo.as_ref()).await
                 };
+                // Persist template_id to workflow
+                if let Some(ref tid) = template_id {
+                    let _ = sqlx::query("UPDATE workflows SET template_id = ? WHERE id = ?")
+                        .bind(tid)
+                        .bind(&workflow_id)
+                        .execute(&pool)
+                        .await;
+                }
+
                 let loaded_template = if let Some(ref tid) = template_id {
                     tracing::info!("Template selected for workflow {}: {}", workflow_id, tid);
                     let _ = tokio::fs::create_dir_all(&workdir).await;
@@ -394,15 +406,35 @@ async fn agent_loop(
                 .ok()
                 .flatten();
 
-                let state = snapshot
+                let mut state = snapshot
                     .and_then(|json| serde_json::from_str::<AgentState>(&json).ok())
                     .unwrap_or_else(AgentState::new);
 
-                // Process feedback: LLM decides whether to revise plan
+                // For failed/done workflows: reset failed steps and continue directly
-                let state = process_feedback(
+                // For running workflows: process feedback via LLM
+                let is_resuming = wf.status == "failed" || wf.status == "done";
+                if is_resuming {
+                    // Reset Failed steps to Pending so they get re-executed
+                    for step in &mut state.steps {
+                        if matches!(step.status, StepStatus::Failed) {
+                            step.status = StepStatus::Pending;
+                        }
+                    }
+                    // Attach comment as feedback to the first actionable step
+                    if let Some(order) = state.first_actionable_step() {
+                        if let Some(step) = state.steps.iter_mut().find(|s| s.order == order) {
+                            step.user_feedbacks.push(content.clone());
+                        }
+                    }
+                    tracing::info!("[workflow {}] Resuming from state, first actionable: {:?}",
+                        workflow_id, state.first_actionable_step());
+                } else {
+                    // Active workflow: LLM decides whether to revise plan
+                    state = process_feedback(
                         &llm, &pool, &broadcast_tx,
                         &project_id, &workflow_id, state, &content,
                     ).await;
+                }
 
                 // If there are actionable steps, resume execution
                 if state.first_actionable_step().is_some() {
@@ -418,7 +450,6 @@ async fn agent_loop(
                         .await;
 
                     // Prepare state for execution: set first pending step to Running
-                    let mut state = state;
                     if let Some(next) = state.first_actionable_step() {
                         if let Some(step) = state.steps.iter_mut().find(|s| s.order == next) {
                             if matches!(step.status, StepStatus::Pending) {
@@ -431,12 +462,31 @@ async fn agent_loop(
 
                     let instructions = read_instructions(&workdir).await;
 
-                    // Try to detect which template was used (check for tools/ in workdir parent template)
+                    // Reload external tools from template if available
-                    // For comments, we don't re-load the template — external tools are not available in feedback resume
+                    let ext_tools = if !wf.template_id.is_empty() {
+                        let tid = &wf.template_id;
+                        if template::is_repo_template(tid) {
+                            match template::extract_repo_template(tid, mgr.template_repo.as_ref()).await {
+                                Ok(template_dir) => {
+                                    LoadedTemplate::load_from_dir(tid, &template_dir).await
+                                        .ok().map(|t| t.external_tools)
+                                }
+                                Err(e) => {
+                                    tracing::warn!("Failed to reload template {}: {}", tid, e);
+                                    None
+                                }
+                            }
+                        } else {
+                            LoadedTemplate::load(tid).await.ok().map(|t| t.external_tools)
+                        }
+                    } else {
+                        None
+                    };
+
                     let result = run_agent_loop(
                         &llm, &exec, &pool, &broadcast_tx,
                         &project_id, &workflow_id, &wf.requirement, &workdir, &mgr,
-                        &instructions, Some(state), None, &mut rx,
+                        &instructions, Some(state), ext_tools.as_ref(), &mut rx,
                     ).await;
 
                     let final_status = if result.is_ok() { "done" } else { "failed" };
@@ -648,10 +698,24 @@ fn build_step_tools() -> Vec<Tool> {
             },
             "required": ["reason"]
         })),
-        make_tool("step_done", "完成当前步骤。必须提供摘要，概括本步骤做了什么、结果如何。", serde_json::json!({
+        make_tool("step_done", "完成当前步骤。必须提供摘要。可选声明本步骤的产出物。", serde_json::json!({
             "type": "object",
             "properties": {
-                "summary": { "type": "string", "description": "本步骤的工作摘要" }
+                "summary": { "type": "string", "description": "本步骤的工作摘要" },
+                "artifacts": {
+                    "type": "array",
+                    "description": "本步骤的产出物列表",
+                    "items": {
+                        "type": "object",
+                        "properties": {
+                            "name": { "type": "string", "description": "产物名称" },
+                            "path": { "type": "string", "description": "文件路径（相对工作目录）" },
+                            "type": { "type": "string", "enum": ["file", "json", "markdown"], "description": "产物类型" },
+                            "description": { "type": "string", "description": "一句话说明" }
+                        },
+                        "required": ["name", "path", "type"]
+                    }
+                }
             },
             "required": ["summary"]
         })),
@@ -688,7 +752,7 @@ fn build_step_execution_prompt(project_id: &str, instructions: &str) -> String {
 }
 
 /// Build user message for a step sub-loop
-fn build_step_user_message(step: &Step, completed_summaries: &[(i32, String, String)], parent_scratchpad: &str) -> String {
+fn build_step_user_message(step: &Step, completed_summaries: &[(i32, String, String, Vec<Artifact>)], parent_scratchpad: &str) -> String {
     let mut ctx = String::new();
 
     ctx.push_str(&format!("## 当前步骤（步骤 {}）\n", step.order));
@@ -705,8 +769,14 @@ fn build_step_user_message(step: &Step, completed_summaries: &[(i32, String, Str
 
     if !completed_summaries.is_empty() {
         ctx.push_str("## 已完成步骤摘要\n");
-        for (order, title, summary) in completed_summaries {
+        for (order, title, summary, artifacts) in completed_summaries {
             ctx.push_str(&format!("- 步骤 {} ({}): {}\n", order, title, summary));
+            if !artifacts.is_empty() {
+                let arts: Vec<String> = artifacts.iter()
+                    .map(|a| format!("{} ({})", a.name, a.artifact_type))
+                    .collect();
+                ctx.push_str(&format!("  产物: {}\n", arts.join(", ")));
+            }
         }
         ctx.push('\n');
     }
@@ -1022,6 +1092,7 @@ async fn process_feedback(
                         summary: None,
                         user_feedbacks: Vec::new(),
                         db_id: String::new(),
+                        artifacts: Vec::new(),
                     }
                 }).collect();
 
@@ -1071,7 +1142,7 @@ async fn run_step_loop(
     mgr: &Arc<AgentManager>,
     instructions: &str,
     step: &Step,
-    completed_summaries: &[(i32, String, String)],
+    completed_summaries: &[(i32, String, String, Vec<Artifact>)],
     parent_scratchpad: &str,
     external_tools: Option<&ExternalToolManager>,
     event_rx: &mut mpsc::Receiver<AgentEvent>,
@@ -1121,6 +1192,7 @@ async fn run_step_loop(
                 tracing::error!("[workflow {}] Step {} LLM call failed: {}", workflow_id, step_order, e);
                 return StepResult {
                     status: StepResultStatus::Failed { error: format!("LLM call failed: {}", e) },
+                    artifacts: Vec::new(),
                     summary: format!("LLM 调用失败: {}", e),
                 };
             }
@@ -1137,6 +1209,7 @@ async fn run_step_loop(
                 return StepResult {
                     status: StepResultStatus::Failed { error: "No response from LLM".into() },
                     summary: "LLM 无响应".into(),
+                    artifacts: Vec::new(),
                 };
             }
         };
@@ -1162,11 +1235,41 @@ async fn run_step_loop(
                 match tc.function.name.as_str() {
                     "step_done" => {
                         let summary = args["summary"].as_str().unwrap_or("").to_string();
+                        let artifacts: Vec<Artifact> = args.get("artifacts")
+                            .and_then(|v| v.as_array())
+                            .map(|arr| arr.iter().filter_map(|a| {
+                                Some(Artifact {
+                                    name: a["name"].as_str()?.to_string(),
+                                    path: a["path"].as_str()?.to_string(),
+                                    artifact_type: a["type"].as_str().unwrap_or("file").to_string(),
+                                    description: a["description"].as_str().unwrap_or("").to_string(),
+                                })
+                            }).collect())
+                            .unwrap_or_default();
+
+                        // Save artifacts to DB
+                        for art in &artifacts {
+                            let art_id = uuid::Uuid::new_v4().to_string();
+                            let _ = sqlx::query(
+                                "INSERT INTO step_artifacts (id, workflow_id, step_order, name, path, artifact_type, description) VALUES (?, ?, ?, ?, ?, ?, ?)"
+                            )
+                            .bind(&art_id)
+                            .bind(workflow_id)
+                            .bind(step_order)
+                            .bind(&art.name)
+                            .bind(&art.path)
+                            .bind(&art.artifact_type)
+                            .bind(&art.description)
+                            .execute(pool)
+                            .await;
+                        }
+
                         log_execution(pool, broadcast_tx, workflow_id, step_order, "step_done", &summary, "步骤完成", "done").await;
                         step_chat_history.push(ChatMessage::tool_result(&tc.id, "步骤已完成。"));
                         step_done_result = Some(StepResult {
                             status: StepResultStatus::Done,
                             summary,
+                            artifacts,
                         });
                     }
 
@@ -1222,6 +1325,7 @@ async fn run_step_loop(
                                     return StepResult {
                                         status: StepResultStatus::Failed { error: "Event channel closed".into() },
                                         summary: "事件通道关闭".into(),
+                                        artifacts: Vec::new(),
                                     };
                                 }
                             }
@@ -1236,6 +1340,7 @@ async fn run_step_loop(
                             step_done_result = Some(StepResult {
                                 status: StepResultStatus::Failed { error: format!("用户终止: {}", reason) },
                                 summary: format!("用户终止了执行: {}", reason),
+                                artifacts: Vec::new(),
                             });
                             continue;
                         }
@@ -1437,6 +1542,7 @@ async fn run_step_loop(
     StepResult {
         status: StepResultStatus::Failed { error: "步骤迭代次数超限（50轮）".into() },
         summary: "步骤执行超过50轮迭代限制，未能完成".into(),
+        artifacts: Vec::new(),
     }
 }
 
@@ -1478,6 +1584,7 @@ async fn plan_infos_from_state_with_override(
                     description: s.title.clone(),
                     command: s.description.clone(),
                     status: Some(status),
+                    artifacts: s.artifacts.clone(),
                 }
             }).collect();
         }
@@ -1567,6 +1674,7 @@ async fn run_agent_loop(
                                 order, title, description: detail,
                                 status: StepStatus::Pending, summary: None,
                                 user_feedbacks: Vec::new(), db_id: String::new(),
+                                artifacts: Vec::new(),
                             });
                         }
 
@@ -1633,9 +1741,9 @@ async fn run_agent_loop(
         save_state_snapshot(pool, workflow_id, step_order, &state).await;
 
         // Build completed summaries for context
-        let completed_summaries: Vec<(i32, String, String)> = state.steps.iter()
+        let completed_summaries: Vec<(i32, String, String, Vec<Artifact>)> = state.steps.iter()
             .filter(|s| matches!(s.status, StepStatus::Done))
-            .map(|s| (s.order, s.title.clone(), s.summary.clone().unwrap_or_default()))
+            .map(|s| (s.order, s.title.clone(), s.summary.clone().unwrap_or_default(), s.artifacts.clone()))
             .collect();
 
         let step = state.steps.iter().find(|s| s.order == step_order).unwrap().clone();
@@ -1658,6 +1766,7 @@ async fn run_agent_loop(
                 if let Some(s) = state.steps.iter_mut().find(|s| s.order == step_order) {
                     s.status = StepStatus::Done;
                     s.summary = Some(step_result.summary.clone());
+                    s.artifacts = step_result.artifacts.clone();
                 }
             }
             StepResultStatus::Failed { error } => {
@@ -1769,6 +1878,7 @@ async fn run_agent_loop(
                                     description: item["description"].as_str().unwrap_or("").to_string(),
                                     status: StepStatus::Pending, summary: None,
                                     user_feedbacks: Vec::new(), db_id: String::new(),
+                                    artifacts: Vec::new(),
                                 }
                             }).collect();
 
@@ -1889,6 +1999,7 @@ mod tests {
             summary: None,
             user_feedbacks: Vec::new(),
             db_id: String::new(),
+            artifacts: Vec::new(),
         }
     }
 
@@ -1911,8 +2022,8 @@ mod tests {
     fn step_msg_with_completed_summaries() {
         let step = make_step(3, "Deploy", "Push to prod", StepStatus::Running);
         let summaries = vec![
-            (1, "Setup".into(), "Installed deps".into()),
+            (1, "Setup".into(), "Installed deps".into(), Vec::new()),
-            (2, "Build".into(), "Compiled OK".into()),
+            (2, "Build".into(), "Compiled OK".into(), Vec::new()),
         ];
         let msg = build_step_user_message(&step, &summaries, "");
 
@@ -1951,8 +2062,8 @@ mod tests {
             ..make_step(3, "API", "build REST API", StepStatus::Running)
         };
         let summaries = vec![
-            (1, "DB".into(), "Schema created".into()),
+            (1, "DB".into(), "Schema created".into(), Vec::new()),
-            (2, "Models".into(), "ORM models done".into()),
+            (2, "Models".into(), "ORM models done".into(), Vec::new()),
         ];
         let msg = build_step_user_message(&step, &summaries, "tech_stack=FastAPI");
 

src/db.rs

@@ -59,6 +59,13 @@ impl Database {
         .execute(&self.pool)
         .await;
 
+        // Migration: add template_id column to workflows
+        let _ = sqlx::query(
+            "ALTER TABLE workflows ADD COLUMN template_id TEXT NOT NULL DEFAULT ''"
+        )
+        .execute(&self.pool)
+        .await;
+
         // Migration: add deleted column to projects
         let _ = sqlx::query(
             "ALTER TABLE projects ADD COLUMN deleted INTEGER NOT NULL DEFAULT 0"
@@ -208,6 +215,20 @@ impl Database {
         .execute(&self.pool)
         .await?;
 
+        sqlx::query(
+            "CREATE TABLE IF NOT EXISTS step_artifacts (
+                id TEXT PRIMARY KEY,
+                workflow_id TEXT NOT NULL,
+                step_order INTEGER NOT NULL,
+                name TEXT NOT NULL,
+                path TEXT NOT NULL,
+                artifact_type TEXT NOT NULL DEFAULT 'file',
+                description TEXT NOT NULL DEFAULT ''
+            )"
+        )
+        .execute(&self.pool)
+        .await?;
+
         Ok(())
     }
 }
@@ -231,6 +252,7 @@ pub struct Workflow {
     pub status: String,
     pub created_at: String,
     pub report: String,
+    pub template_id: String,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize, sqlx::FromRow)]

src/llm.rs

@@ -93,7 +93,11 @@ pub struct ChatChoice {
 impl LlmClient {
     pub fn new(config: &LlmConfig) -> Self {
         Self {
-            client: reqwest::Client::new(),
+            client: reqwest::Client::builder()
+                .timeout(std::time::Duration::from_secs(120))
+                .connect_timeout(std::time::Duration::from_secs(10))
+                .build()
+                .expect("Failed to build HTTP client"),
             config: config.clone(),
         }
     }
@@ -106,22 +110,50 @@ impl LlmClient {
             .unwrap_or_default())
     }
 
-    /// Chat with tool definitions — returns full response for tool-calling loop
+    /// Chat with tool definitions — returns full response for tool-calling loop.
+    /// Retries up to 3 times with exponential backoff on transient errors.
     pub async fn chat_with_tools(&self, messages: Vec<ChatMessage>, tools: &[Tool]) -> anyhow::Result<ChatResponse> {
         let url = format!("{}/chat/completions", self.config.base_url);
-        tracing::debug!("LLM request to {} model={} messages={} tools={}", url, self.config.model, messages.len(), tools.len());
+        let max_retries = 3u32;
-        let http_resp = self.client
+        let mut last_err = None;
+        let tools_vec = tools.to_vec();
+
+        for attempt in 0..max_retries {
+            if attempt > 0 {
+                let delay = std::time::Duration::from_secs(2u64.pow(attempt));
+                tracing::warn!("LLM retry #{} after {}s", attempt, delay.as_secs());
+                tokio::time::sleep(delay).await;
+            }
+
+            tracing::debug!("LLM request to {} model={} messages={} tools={} attempt={}", url, self.config.model, messages.len(), tools_vec.len(), attempt + 1);
+            let result = self.client
                 .post(&url)
                 .header("Authorization", format!("Bearer {}", self.config.api_key))
                 .json(&ChatRequest {
                     model: self.config.model.clone(),
-                messages,
+                    messages: messages.clone(),
-                tools: tools.to_vec(),
+                    tools: tools_vec.clone(),
                 })
                 .send()
-            .await?;
+                .await;
+
+            let http_resp = match result {
+                Ok(r) => r,
+                Err(e) => {
+                    tracing::warn!("LLM request error (attempt {}): {}", attempt + 1, e);
+                    last_err = Some(anyhow::anyhow!("{}", e));
+                    continue;
+                }
+            };
 
             let status = http_resp.status();
+            if status.is_server_error() || status.as_u16() == 429 {
+                let body = http_resp.text().await.unwrap_or_default();
+                tracing::warn!("LLM API error {} (attempt {}): {}", status, attempt + 1, &body[..body.len().min(200)]);
+                last_err = Some(anyhow::anyhow!("LLM API error {}: {}", status, body));
+                continue;
+            }
+
             if !status.is_success() {
                 let body = http_resp.text().await.unwrap_or_default();
                 tracing::error!("LLM API error {}: {}", status, &body[..body.len().min(500)]);
@@ -134,6 +166,9 @@ impl LlmClient {
                 anyhow::anyhow!("Failed to parse LLM response: {}", e)
             })?;
 
-        Ok(resp)
+            return Ok(resp);
+        }
+
+        Err(last_err.unwrap_or_else(|| anyhow::anyhow!("LLM call failed after {} retries", max_retries)))
     }
 }

src/prompts/step_execution.md

@@ -20,6 +20,7 @@
 
 - **专注当前步骤**，不做超出范围的事
 - 完成后**必须**调用 step_done(summary)，summary 应简洁概括本步骤做了什么、结果如何
+- 完成步骤时，用 `step_done` 的 `artifacts` 参数声明本步骤产出的文件。每个产出物需要 name、path、type (file/json/markdown)
 - 需要用户确认时使用 wait_for_approval(reason)
 - update_scratchpad 用于记录本步骤内的中间状态，是工作记忆而非日志，只保留当前有用的信息
 

src/state.rs

@@ -8,6 +8,7 @@ use crate::llm::ChatMessage;
 pub struct StepResult {
     pub status: StepResultStatus,
     pub summary: String,
+    pub artifacts: Vec<Artifact>,
 }
 
 #[derive(Debug, Clone)]
@@ -54,6 +55,15 @@ pub enum StepStatus {
     Failed,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Artifact {
+    pub name: String,
+    pub path: String,
+    pub artifact_type: String,
+    #[serde(default)]
+    pub description: String,
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct Step {
     pub order: i32,
@@ -68,6 +78,9 @@ pub struct Step {
     pub user_feedbacks: Vec<String>,
     #[serde(default)]
     pub db_id: String,
+    /// 步骤产出物
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub artifacts: Vec<Artifact>,
 }
 
 // --- Core state ---
@@ -203,6 +216,12 @@ impl AgentState {
             for s in done {
                 let summary = s.summary.as_deref().unwrap_or("(no summary)");
                 ctx.push_str(&format!("- 步骤 {}: {}\n", s.order, summary));
+                if !s.artifacts.is_empty() {
+                    let arts: Vec<String> = s.artifacts.iter()
+                        .map(|a| format!("{} ({})", a.name, a.artifact_type))
+                        .collect();
+                    ctx.push_str(&format!("  产物: {}\n", arts.join(", ")));
+                }
             }
             ctx.push('\n');
         }
@@ -249,6 +268,7 @@ mod tests {
             summary: None,
             user_feedbacks: Vec::new(),
             db_id: String::new(),
+            artifacts: Vec::new(),
         }
     }
 

web/src/components/PlanSection.vue

@@ -59,6 +59,11 @@ function quoteStep(e: Event, step: PlanStepInfo) {
         <div v-if="step.command && expandedSteps.has(step.order)" class="step-detail">
           {{ step.command }}
         </div>
+        <div v-if="step.artifacts?.length" class="step-artifacts">
+          <span v-for="a in step.artifacts" :key="a.path" class="artifact-tag">
+            📄 {{ a.name }} <span class="artifact-type">{{ a.artifact_type }}</span>
+          </span>
+        </div>
       </div>
       <div v-if="!steps.length" class="empty-state">
         AI 将在这里展示执行计划
@@ -188,6 +193,30 @@ function quoteStep(e: Event, step: PlanStepInfo) {
   border-top: 1px solid var(--border);
 }
 
+.step-artifacts {
+  padding: 4px 10px 8px 44px;
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+}
+
+.artifact-tag {
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  font-size: 11px;
+  color: var(--text-secondary);
+  background: var(--bg-tertiary);
+  padding: 2px 8px;
+  border-radius: 4px;
+}
+
+.artifact-type {
+  font-size: 10px;
+  color: var(--accent);
+  opacity: 0.8;
+}
+
 .empty-state {
   color: var(--text-secondary);
   font-size: 13px;

web/src/types.ts

@@ -26,11 +26,19 @@ export interface ExecutionLogEntry {
   created_at: string
 }
 
+export interface StepArtifact {
+  name: string
+  path: string
+  artifact_type: string
+  description: string
+}
+
 export interface PlanStepInfo {
   order: number
   description: string
   command: string
   status?: 'pending' | 'running' | 'done' | 'failed'
+  artifacts?: StepArtifact[]
 }
 
 export interface Comment {