bamboo_engine/runtime/execution/
agent_spawn.rs

1//! Core agent execution spawning logic.
2//!
3//! Provides [`spawn_session_execution`] which handles the full lifecycle of a
4//! background agent run: spawn task → execute → finalize runner → persist session.
5
6use std::collections::{BTreeSet, HashMap};
7use std::sync::Arc;
8
9use tokio::sync::{mpsc, RwLock};
10use tokio_util::sync::CancellationToken;
11use tracing::Instrument;
12
13use bamboo_agent_core::tools::ToolExecutor;
14use bamboo_agent_core::{AgentEvent, Session};
15use bamboo_domain::ReasoningEffort;
16use bamboo_infrastructure::LLMProvider;
17
18use crate::runtime::config::ImageFallbackConfig;
19use crate::runtime::execution::runner_lifecycle::finalize_runner;
20use crate::runtime::execution::runner_state::AgentRunner;
21use crate::runtime::Agent;
22use crate::runtime::ExecuteRequest;
23
24const SKILL_CONTEXT_START_MARKER: &str = "<!-- BAMBOO_SKILL_CONTEXT_START -->";
25const TOOL_GUIDE_START_MARKER: &str = "<!-- BAMBOO_TOOL_GUIDE_START -->";
26const EXTERNAL_MEMORY_START_MARKER: &str = "<!-- BAMBOO_EXTERNAL_MEMORY_START -->";
27const TASK_LIST_START_MARKER: &str = "<!-- BAMBOO_TASK_LIST_START -->";
28
29/// Arguments for spawning a background agent execution.
30///
31/// This is the crate-agnostic equivalent of the server's `SpawnAgentExecution`.
32/// It holds everything needed to run the agent loop and persist the result,
33/// without depending on HTTP types or `AppState`.
34pub struct SessionExecutionArgs {
35    // Core execution.
36    pub agent: Arc<Agent>,
37    pub session_id: String,
38    pub session: Session,
39
40    // Execution parameters.
41    pub tools_override: Option<Arc<dyn ToolExecutor>>,
42    pub provider_override: Option<Arc<dyn LLMProvider>>,
43    pub provider_name: Option<String>,
44    pub provider_type: Option<String>,
45    pub model: String,
46    pub fast_model: Option<String>,
47    /// Optional provider override for lightweight fast-model calls.
48    pub fast_model_provider: Option<Arc<dyn LLMProvider>>,
49    pub background_model: Option<String>,
50    /// Optional provider override for memory/background model calls.
51    pub background_model_provider: Option<Arc<dyn LLMProvider>>,
52    pub summarization_model: Option<String>,
53    /// Optional provider override for summarization / context compression calls.
54    pub summarization_model_provider: Option<Arc<dyn LLMProvider>>,
55    pub reasoning_effort: Option<ReasoningEffort>,
56    pub reasoning_effort_source: String,
57    pub disabled_tools: Option<BTreeSet<String>>,
58    pub disabled_skill_ids: Option<BTreeSet<String>>,
59    pub selected_skill_ids: Option<Vec<String>>,
60    pub selected_skill_mode: Option<String>,
61    pub cancel_token: CancellationToken,
62    pub mpsc_tx: mpsc::Sender<AgentEvent>,
63    pub image_fallback: Option<ImageFallbackConfig>,
64    pub app_data_dir: Option<std::path::PathBuf>,
65
66    // Post-execution resources.
67    pub runners: Arc<RwLock<HashMap<String, AgentRunner>>>,
68    pub sessions_cache: Arc<RwLock<HashMap<String, Session>>>,
69}
70
71/// Spawn a background agent execution task.
72///
73/// This function spawns a tokio task that:
74/// 1. Executes the agent loop via `agent.execute()`
75/// 2. Sends a terminal error event if the execution fails
76/// 3. Finalizes the runner status
77/// 4. Persists the session via merge-save (preserves concurrent UI title/pin edits)
78/// 5. Updates the in-memory session cache
79pub fn spawn_session_execution(args: SessionExecutionArgs) {
80    let span_session_id = args.session_id.clone();
81    let session_span = tracing::info_span!("agent_execution", session_id = %span_session_id);
82
83    tokio::spawn(
84        async move {
85            let SessionExecutionArgs {
86                agent,
87                session_id,
88                mut session,
89                tools_override,
90                provider_override,
91                provider_name,
92                provider_type,
93                model,
94                fast_model,
95                fast_model_provider,
96                background_model,
97                background_model_provider,
98                summarization_model,
99                summarization_model_provider,
100                reasoning_effort,
101                reasoning_effort_source,
102                disabled_tools,
103                disabled_skill_ids,
104                selected_skill_ids,
105                selected_skill_mode,
106                cancel_token,
107                mpsc_tx,
108                image_fallback,
109                app_data_dir,
110                runners,
111                sessions_cache,
112            } = args;
113
114            let initial_message = initial_user_message_for_session(&session);
115            let selected_skill_ids =
116                selected_skill_ids.or_else(|| selected_skill_ids_for_session(&session));
117            let selected_skill_mode =
118                selected_skill_mode.or_else(|| selected_skill_mode_for_session(&session));
119
120            tracing::info!(
121                "[{}] Using resolved session model: {}, reasoning_effort={}, reasoning_source={}",
122                session_id,
123                model,
124                reasoning_effort
125                    .map(ReasoningEffort::as_str)
126                    .unwrap_or("none"),
127                reasoning_effort_source
128            );
129
130            session.model = model.clone();
131
132            let system_prompt = system_prompt_for_session(&session);
133            if let Some(prompt) = system_prompt.as_ref() {
134                log_base_system_prompt_snapshot(&session_id, prompt);
135            }
136
137            let result = agent
138                .execute(
139                    &mut session,
140                    ExecuteRequest {
141                        initial_message,
142                        event_tx: mpsc_tx.clone(),
143                        cancel_token,
144                        tools: tools_override,
145                        provider_override,
146                        model: Some(model),
147                        provider_name,
148                        provider_type,
149                        fast_model,
150                        fast_model_provider,
151                        background_model,
152                        background_model_provider,
153                        summarization_model,
154                        summarization_model_provider,
155                        reasoning_effort,
156                        disabled_tools,
157                        disabled_skill_ids,
158                        selected_skill_ids,
159                        selected_skill_mode,
160                        image_fallback,
161                        app_data_dir,
162                    },
163                )
164                .await;
165
166            // Send terminal event for all error cases (including cancellation).
167            if let Some(error_event) = terminal_error_event_for_result(&result) {
168                let _ = mpsc_tx.send(error_event).await;
169            }
170
171            // Update runner status.
172            finalize_runner(&runners, &session_id, &result).await;
173
174            // Save session via merge-save so any concurrent UI edits to
175            // title / pinned / title_version are preserved (the runtime is not
176            // an authoritative title writer).
177            if let Err(error) = agent.persistence().save_runtime_session(&mut session).await {
178                tracing::warn!("[{}] Failed to save session: {}", session_id, error);
179            }
180
181            // Update memory cache.
182            {
183                let mut sessions = sessions_cache.write().await;
184                sessions.insert(session_id.clone(), session);
185            }
186
187            tracing::info!("[{}] Agent execution completed", session_id);
188        }
189        .instrument(session_span),
190    );
191}
192
193/// Log a snapshot of the base system prompt for debugging.
194pub fn log_base_system_prompt_snapshot(session_id: &str, prompt: &str) {
195    tracing::info!(
196        "[{}] Base system prompt snapshot: len={} chars, has_skill={}, has_tool_guide={}, has_external_memory={}, has_task_list={}",
197        session_id,
198        prompt.len(),
199        prompt.contains(SKILL_CONTEXT_START_MARKER),
200        prompt.contains(TOOL_GUIDE_START_MARKER),
201        prompt.contains(EXTERNAL_MEMORY_START_MARKER),
202        prompt.contains(TASK_LIST_START_MARKER),
203    );
204
205    tracing::debug!(
206        "[{}] ========== BASE SYSTEM PROMPT SNAPSHOT ==========",
207        session_id
208    );
209    tracing::debug!("[{}] Snapshot length: {} chars", session_id, prompt.len());
210    tracing::debug!("[{}] -----------------------------------", session_id);
211    tracing::debug!("[{}] {}", session_id, prompt);
212    tracing::debug!(
213        "[{}] ========== END BASE SYSTEM PROMPT SNAPSHOT ==========",
214        session_id
215    );
216}
217
218/// Map an execution result to a terminal error event.
219pub fn terminal_error_event_for_result<E>(result: &Result<(), E>) -> Option<AgentEvent>
220where
221    E: std::fmt::Display,
222{
223    match result {
224        Ok(_) => None,
225        Err(error) if is_cancelled_error(error) => Some(AgentEvent::Error {
226            message: "Agent execution cancelled by user".to_string(),
227        }),
228        Err(error) => Some(AgentEvent::Error {
229            message: error.to_string(),
230        }),
231    }
232}
233
234fn is_cancelled_error<E>(error: &E) -> bool
235where
236    E: std::fmt::Display,
237{
238    error.to_string().contains("cancelled")
239}
240
241// Session metadata helpers (pure functions, no server dependency).
242
243fn system_prompt_for_session(session: &Session) -> Option<String> {
244    session
245        .messages
246        .iter()
247        .find(|message| matches!(message.role, bamboo_agent_core::Role::System))
248        .map(|message| message.content.clone())
249}
250
251fn initial_user_message_for_session(session: &Session) -> String {
252    session
253        .messages
254        .last()
255        .filter(|message| matches!(message.role, bamboo_agent_core::Role::User))
256        .map(|message| message.content.clone())
257        .unwrap_or_default()
258}
259
260fn selected_skill_ids_for_session(session: &Session) -> Option<Vec<String>> {
261    session
262        .metadata
263        .get("selected_skill_ids")
264        .and_then(|raw| crate::skills::selection::parse_selected_skill_ids_metadata(raw))
265}
266
267fn selected_skill_mode_for_session(session: &Session) -> Option<String> {
268    let value = session
269        .metadata
270        .get("skill_mode")
271        .or_else(|| session.metadata.get("mode"))?;
272    let trimmed = value.trim();
273    if trimmed.is_empty() {
274        None
275    } else {
276        Some(trimmed.to_string())
277    }
278}
bamboo_engine/runtime/execution/agent_spawn.rs

bamboo_engine/runtime/execution/
agent_spawn.rs
