Add sandbox

crates/herald-server/src/bot_actions/review.rs

@@ -1,22 +1,33 @@
 use futures_util::stream::TryStreamExt;
+use openrouter_rs::types::Tool;
 use tokio::io::AsyncReadExt;
 use tokio_util::io::StreamReader;
-use tracing::instrument;
+use tracing::{info, instrument, warn};
 
 use crate::{
     bot::ReviewResult,
-    consts::{BOT_PROCESS_MSG, MAX_DIFF_SIZE, REVIEW_PROMPT},
+    consts::{BOT_PROCESS_MSG, MAX_DIFF_SIZE, REVIEW_PROMPT, SANDBOX_SYSTEM_PROMPT},
     gitea::{GiteaAPI, ReviewPayload},
     metrics,
     open_router::OpenRouterClient,
+    sandbox::{Sandbox, SandboxConfig, agent},
 };
 
-#[instrument(skip(gitea_api, open_router_client, http_client, review_payload))]
+#[instrument(skip(
+    gitea_api,
+    open_router_client,
+    http_client,
+    sandbox_config,
+    tools,
+    review_payload
+))]
 pub async fn exec_review(
     gitea_api: &GiteaAPI,
     open_router_client: &OpenRouterClient,
     http_client: &reqwest::Client,
     model: &str,
+    sandbox_config: &SandboxConfig,
+    tools: Vec<Tool>,
     review_payload: ReviewPayload,
 ) -> anyhow::Result<()> {
     tracing::info!(
@@ -45,10 +56,24 @@ pub async fn exec_review(
             .replace("{comment}", &review_payload.comment.body)
             .replace("{diff}", &diff_for_llm);
 
-        let chat_result = open_router_client.chat(&bot_request).await?;
-        let mut review_result = serde_json::from_str::<ReviewResult>(&chat_result.message)?;
+        let (message, cost) = if sandbox_config.enabled {
+            run_sandboxed_review(
+                gitea_api,
+                open_router_client,
+                sandbox_config,
+                tools,
+                &review_payload,
+                &bot_request,
+            )
+            .await?
+        } else {
+            let chat_result = open_router_client.chat(&bot_request).await?;
+            (chat_result.message, chat_result.cost)
+        };
 
-        review_result.cost = chat_result.cost;
+        let mut review_result = serde_json::from_str::<ReviewResult>(&message)?;
+
+        review_result.cost = cost;
         if let Some(cost) = review_result.cost {
             metrics::openrouter_cost_usd(cost);
         }
@@ -86,6 +111,50 @@ pub async fn exec_review(
     }
 }
 
+/// Runs the review inside a sandbox container, letting the model explore the
+/// repository with tools before answering.
+async fn run_sandboxed_review(
+    gitea_api: &GiteaAPI,
+    open_router_client: &OpenRouterClient,
+    sandbox_config: &SandboxConfig,
+    tools: Vec<Tool>,
+    review_payload: &ReviewPayload,
+    bot_request: &str,
+) -> anyhow::Result<(String, Option<f64>)> {
+    let repo_url = gitea_api.repo_clone_url(&review_payload.repository.full_name);
+
+    let sandbox = Sandbox::create(
+        &sandbox_config.runtime,
+        &repo_url,
+        gitea_api.token(),
+        review_payload.pull_request.number,
+    )
+    .await?;
+
+    let result = agent::run(
+        open_router_client,
+        &sandbox,
+        tools,
+        SANDBOX_SYSTEM_PROMPT,
+        bot_request,
+        sandbox_config.max_iterations,
+    )
+    .await;
+
+    if let Err(err) = sandbox.cleanup().await {
+        warn!(%err, "Failed to clean up sandbox container");
+    }
+
+    let result = result?;
+    info!(
+        iterations = result.iterations,
+        cost = ?result.cost,
+        "Sandboxed review finished"
+    );
+
+    Ok((result.message, result.cost))
+}
+
 fn review_result_to_markdown(review_result: &ReviewResult) -> String {
     if review_result.reviews.is_empty() {
         return String::from("No issues found. ✅");