zeptoclaw 0.9.0

Ultra-lightweight personal AI assistant
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
//! Agent delegation tool for multi-agent swarms.
//!
//! The `DelegateTool` creates a temporary `AgentLoop` with a role-specific
//! system prompt and tool whitelist, runs it to completion, and returns
//! the result to the calling (lead) agent.

use std::sync::Arc;

use async_trait::async_trait;
use serde_json::{json, Value};
use tokio::sync::Semaphore;
use tracing::{info, warn};

use futures::future::join_all;

use crate::agent::{AgentLoop, ContextBuilder, SwarmScratchpad};
use crate::bus::{InboundMessage, MessageBus};
use crate::config::Config;
use crate::error::{Result, ZeptoError};
use crate::providers::{ChatOptions, LLMProvider, LLMResponse, ToolDefinition};
use crate::runtime::NativeRuntime;
use crate::session::{Message, SessionManager};
use crate::tools::filesystem::{EditFileTool, ListDirTool, ReadFileTool, WriteFileTool};
use crate::tools::memory::{MemoryGetTool, MemorySearchTool};
use crate::tools::message::MessageTool;
use crate::tools::shell::ShellTool;
use crate::tools::web::WebFetchTool;
use crate::tools::EchoTool;

use super::{Tool, ToolCategory, ToolContext, ToolOutput};

/// Tool to delegate a task to a specialist sub-agent.
///
/// Creates a new `AgentLoop` with a role-specific system prompt and optional
/// tool whitelist, runs it to completion, and returns the result. Sub-agents
/// are prevented from delegating further to avoid recursion.
///
/// Supports two actions:
/// - `run` (default) — delegates a single task to one sub-agent.
/// - `aggregate` — fans out multiple tasks (each with its own role) and merges
///   the results using a configurable merge strategy.
///
/// Concurrency is bounded by a `Semaphore` whose capacity comes from
/// `config.swarm.max_concurrent`.
pub struct DelegateTool {
    config: Config,
    provider: Arc<dyn LLMProvider>,
    bus: Arc<MessageBus>,
    /// Semaphore limiting concurrent sub-agent executions.
    semaphore: Arc<Semaphore>,
    /// Shared scratchpad for passing context between sub-agents in a swarm session.
    scratchpad: SwarmScratchpad,
}

impl DelegateTool {
    /// Create a new delegate tool.
    ///
    /// # Arguments
    /// * `config` - Agent configuration (cloned for each sub-agent)
    /// * `provider` - Shared LLM provider (wrapped via `ProviderRef`)
    /// * `bus` - Message bus (a fresh bus is created for each sub-agent)
    pub fn new(config: Config, provider: Arc<dyn LLMProvider>, bus: Arc<MessageBus>) -> Self {
        let max_concurrent = config.swarm.max_concurrent as usize;
        // Guard against a zero-capacity semaphore which would deadlock every acquire.
        let capacity = if max_concurrent == 0 {
            1
        } else {
            max_concurrent
        };
        let semaphore = Arc::new(Semaphore::new(capacity));
        Self {
            config,
            provider,
            bus,
            semaphore,
            scratchpad: SwarmScratchpad::new(),
        }
    }

    /// Create a delegate tool with an explicit semaphore.
    ///
    /// Primarily useful in tests where callers want to control the semaphore
    /// capacity independently of `config.swarm.max_concurrent`.
    pub fn with_semaphore(
        config: Config,
        provider: Arc<dyn LLMProvider>,
        bus: Arc<MessageBus>,
        semaphore: Arc<Semaphore>,
    ) -> Self {
        Self {
            config,
            provider,
            bus,
            semaphore,
            scratchpad: SwarmScratchpad::new(),
        }
    }

    /// Return a reference to the shared swarm scratchpad.
    ///
    /// Primarily useful in tests to inspect scratchpad state after delegating.
    pub fn scratchpad(&self) -> &SwarmScratchpad {
        &self.scratchpad
    }

    /// Create a standard set of tools for a sub-agent.
    ///
    /// Always excludes `delegate` and `spawn` to prevent recursion.
    /// If a whitelist is provided, only tools matching those names are included.
    fn create_sub_agent_tools(&self, whitelist: Option<&[String]>) -> Vec<Box<dyn Tool>> {
        let mut all_tools: Vec<Box<dyn Tool>> = vec![
            Box::new(EchoTool),
            Box::new(ReadFileTool),
            Box::new(WriteFileTool),
            Box::new(ListDirTool),
            Box::new(EditFileTool),
            Box::new(ShellTool::with_runtime(Arc::new(NativeRuntime::new()))),
            Box::new(WebFetchTool::new()),
            Box::new(MessageTool::new(self.bus.clone())),
        ];

        // Add memory tools if enabled
        match &self.config.memory.backend {
            crate::config::MemoryBackend::Disabled => {}
            _ => {
                all_tools.push(Box::new(MemorySearchTool::new(self.config.memory.clone())));
                all_tools.push(Box::new(MemoryGetTool::new(self.config.memory.clone())));
            }
        }

        match whitelist {
            Some(names) => all_tools
                .into_iter()
                .filter(|t| names.iter().any(|n| n == t.name()))
                .collect(),
            None => all_tools,
        }
    }

    /// Run a single delegated sub-agent and return its raw result string.
    ///
    /// This is the shared implementation used by both the `run` and `aggregate`
    /// actions. It acquires a semaphore permit before creating the sub-agent,
    /// so concurrent calls are bounded by `config.swarm.max_concurrent`.
    ///
    /// When `inject_prior_context` is true, the scratchpad contents are injected
    /// into the sub-agent's system prompt so it can build on previous agents'
    /// outputs (sequential mode). When false, the sub-agent runs independently
    /// without seeing prior outputs (parallel mode).
    ///
    /// The returned string does **not** include the `[role]:` prefix; callers
    /// are responsible for any formatting.
    async fn run_single_delegate(
        &self,
        role: &str,
        task: &str,
        tools: Option<&[String]>,
        _ctx: &ToolContext,
        inject_prior_context: bool,
    ) -> Result<String> {
        let role_lower = role.to_lowercase();
        let role_config = self.config.swarm.roles.get(&role_lower);

        // Build system prompt from role config or generate a default
        let mut system_prompt = match role_config {
            Some(rc) if !rc.system_prompt.is_empty() => rc.system_prompt.clone(),
            _ => format!(
                "You are a specialist with the role: {}. \
                 Complete the task given to you thoroughly and return your findings. \
                 You can send interim updates to the user via the message tool.",
                role
            ),
        };

        // Inject previous agent outputs from the scratchpad so this sub-agent
        // can build on what earlier agents produced (only in sequential mode).
        if inject_prior_context {
            if let Some(context) = self.scratchpad.format_for_prompt().await {
                system_prompt = format!("{}\n\n{}", system_prompt, context);
            }
        }

        // Determine allowed tools: explicit override > role config > all
        let allowed_tool_names: Option<Vec<String>> = tools.map(|t| t.to_vec()).or_else(|| {
            role_config
                .filter(|rc| !rc.tools.is_empty())
                .map(|rc| rc.tools.clone())
        });

        info!(role = %role, task_len = task.len(), "Delegating task to sub-agent");

        // Acquire semaphore permit before creating the sub-agent.
        // The permit is held for the duration of this function and released
        // automatically when `_permit` drops at the end of the scope.
        let _permit = self
            .semaphore
            .acquire()
            .await
            .map_err(|_| ZeptoError::Tool("Swarm semaphore closed".into()))?;

        // Create sub-agent with role-specific context
        let session_manager = SessionManager::new_memory();
        let sub_bus = Arc::new(MessageBus::new());
        let context_builder = ContextBuilder::new().with_system_prompt(&system_prompt);

        let sub_agent = AgentLoop::with_context_builder(
            self.config.clone(),
            session_manager,
            sub_bus,
            context_builder,
        );

        // Set the same LLM provider via the ProviderRef wrapper
        sub_agent
            .set_provider(Box::new(ProviderRef(Arc::clone(&self.provider))))
            .await;

        // Register tools (filtered by whitelist)
        let sub_tools = self.create_sub_agent_tools(allowed_tool_names.as_deref());
        for tool in sub_tools {
            sub_agent.register_tool(tool).await;
        }

        // Create the inbound message for the sub-agent
        let delegate_id = uuid::Uuid::new_v4()
            .to_string()
            .chars()
            .take(8)
            .collect::<String>();
        let inbound = InboundMessage::new(
            "delegate",
            &format!("delegate:{}", delegate_id),
            &format!("delegate:{}", delegate_id),
            task,
        );

        // Run the sub-agent to completion
        match sub_agent.process_message(&inbound).await {
            Ok(result) => {
                info!(role = %role, result_len = result.len(), "Sub-agent completed");
                Ok(result)
            }
            Err(e) => {
                warn!(role = %role, error = %e, "Sub-agent failed");
                Err(ZeptoError::Tool(format!(
                    "Sub-agent '{}' failed: {}",
                    role, e
                )))
            }
        }
    }
}

#[async_trait]
impl Tool for DelegateTool {
    fn name(&self) -> &str {
        "delegate"
    }

    fn description(&self) -> &str {
        "Delegate a task to a specialist sub-agent with a specific role. \
         The sub-agent runs to completion and returns its result. \
         Use this to decompose complex tasks into specialist subtasks. \
         Use action='aggregate' with a 'tasks' array for multiple tasks. \
         IMPORTANT: When using aggregate with multiple tasks, ask the user \
         which execution mode they prefer before running: \
         (1) Parallel — all agents run at the same time, faster but independent \
         (no context sharing between agents). \
         (2) Sequential — agents run one after another, each can build on \
         prior agents' results (slower but coordinated). \
         If the user already specified a preference (e.g. 'run in parallel', \
         'run them together', 'one by one'), respect that directly."
    }

    fn compact_description(&self) -> &str {
        "Delegate agent"
    }

    fn category(&self) -> ToolCategory {
        ToolCategory::Shell
    }

    fn parameters(&self) -> Value {
        json!({
            "type": "object",
            "properties": {
                "action": {
                    "type": "string",
                    "enum": ["run", "aggregate"],
                    "description": "Action to perform. 'run' (default) delegates a single task. \
                                    'aggregate' fans out multiple tasks and merges results."
                },
                "role": {
                    "type": "string",
                    "description": "The specialist role (e.g., 'researcher', 'writer', 'analyst'). \
                                    Required for action='run'."
                },
                "task": {
                    "type": "string",
                    "description": "The task for the sub-agent to complete. Required for action='run'."
                },
                "tools": {
                    "type": "array",
                    "items": { "type": "string" },
                    "description": "Optional tool whitelist for action='run'. \
                                    If omitted, uses role preset or all available tools."
                },
                "tasks": {
                    "type": "array",
                    "description": "For action='aggregate': array of task specs, \
                                    each with 'role', 'task', and optional 'tools'.",
                    "items": {
                        "type": "object",
                        "properties": {
                            "role": { "type": "string" },
                            "task": { "type": "string" },
                            "tools": {
                                "type": "array",
                                "items": { "type": "string" }
                            }
                        },
                        "required": ["role", "task"]
                    }
                },
                "parallel": {
                    "type": "boolean",
                    "description": "For action='aggregate': if true, all sub-agents run concurrently \
                                    (faster, but tasks cannot see each other's results). \
                                    If false (default), tasks run sequentially and each sub-agent \
                                    can build on prior agents' outputs. Use true for independent \
                                    tasks; false for dependent chains."
                },
                "merge_strategy": {
                    "type": "string",
                    "enum": ["concatenate", "summarize"],
                    "description": "For action='aggregate': how to merge results. \
                                    'concatenate' (default) joins results as '[Role]: result'. \
                                    'summarize' produces a structured markdown document."
                }
            },
            "required": []
        })
    }

    async fn execute(&self, args: Value, ctx: &ToolContext) -> Result<ToolOutput> {
        // Block recursion: sub-agents cannot delegate further
        if ctx.channel.as_deref() == Some("delegate") {
            return Err(ZeptoError::Tool(
                "Cannot delegate from within a delegated task (recursion limit)".to_string(),
            ));
        }

        // Check if swarm is enabled
        if !self.config.swarm.enabled {
            return Err(ZeptoError::Tool(
                "Delegation is disabled in configuration".to_string(),
            ));
        }

        // Default action is "run" for backwards compatibility.
        let action = args.get("action").and_then(|v| v.as_str()).unwrap_or("run");

        match action {
            "run" => {
                let role = args
                    .get("role")
                    .and_then(|v| v.as_str())
                    .ok_or_else(|| ZeptoError::Tool("Missing required 'role' argument".into()))?;
                let task = args
                    .get("task")
                    .and_then(|v| v.as_str())
                    .ok_or_else(|| ZeptoError::Tool("Missing required 'task' argument".into()))?;
                let tool_override: Option<Vec<String>> =
                    args.get("tools").and_then(|v| v.as_array()).map(|arr| {
                        arr.iter()
                            .filter_map(|v| v.as_str().map(String::from))
                            .collect()
                    });

                let result = self
                    .run_single_delegate(role, task, tool_override.as_deref(), ctx, true)
                    .await?;
                // Write the result to the scratchpad so subsequent sub-agents can
                // see what this agent produced.
                self.scratchpad.write(role, &result).await;
                // Preserve the original output format: "[role]: result"
                Ok(ToolOutput::user_visible(format!("[{}]: {}", role, result)))
            }

            "aggregate" => {
                let tasks = args
                    .get("tasks")
                    .and_then(Value::as_array)
                    .ok_or_else(|| ZeptoError::Tool("'aggregate' requires 'tasks' array".into()))?;

                let parallel = args
                    .get("parallel")
                    .and_then(Value::as_bool)
                    .unwrap_or(false);

                // Parse all task specs upfront (validates before execution).
                let task_specs: Vec<(String, String, Option<Vec<String>>)> = tasks
                    .iter()
                    .map(|task_spec| {
                        let role = task_spec
                            .get("role")
                            .and_then(Value::as_str)
                            .unwrap_or("assistant")
                            .to_string();
                        let task_text = task_spec
                            .get("task")
                            .and_then(Value::as_str)
                            .ok_or_else(|| {
                                ZeptoError::Tool(
                                    "Each task in aggregate must have 'task' field".into(),
                                )
                            })?
                            .to_string();
                        let tools: Option<Vec<String>> =
                            task_spec.get("tools").and_then(Value::as_array).map(|arr| {
                                arr.iter()
                                    .filter_map(|v| v.as_str().map(String::from))
                                    .collect()
                            });
                        Ok((role, task_text, tools))
                    })
                    .collect::<Result<Vec<_>>>()?;

                let results = if parallel {
                    // Parallel fan-out: all sub-agents run concurrently without
                    // scratchpad context injection. Bounded by the semaphore.
                    info!(
                        count = task_specs.len(),
                        "Parallel aggregate: fanning out sub-agents"
                    );
                    let futures = task_specs.iter().map(|(role, task_text, tools)| {
                        self.run_single_delegate(
                            role,
                            task_text,
                            tools.as_deref(),
                            ctx,
                            false, // no scratchpad injection in parallel mode
                        )
                    });
                    let raw_results = join_all(futures).await;

                    let mut results: Vec<(String, String)> = Vec::new();
                    for (i, res) in raw_results.into_iter().enumerate() {
                        let role = &task_specs[i].0;
                        match res {
                            Ok(output) => {
                                self.scratchpad.write(role, &output).await;
                                results.push((role.clone(), output));
                            }
                            Err(e) => {
                                let err_msg = format!("[error]: {}", e);
                                warn!(role = %role, error = %e, "Parallel sub-agent failed");
                                results.push((role.clone(), err_msg));
                            }
                        }
                    }
                    results
                } else {
                    // Sequential: each sub-agent sees prior outputs via scratchpad.
                    let mut results: Vec<(String, String)> = Vec::new();
                    for (role, task_text, tools) in &task_specs {
                        let result = self
                            .run_single_delegate(
                                role,
                                task_text,
                                tools.as_deref(),
                                ctx,
                                true, // inject scratchpad context in sequential mode
                            )
                            .await?;
                        self.scratchpad.write(role, &result).await;
                        results.push((role.clone(), result));
                    }
                    results
                };

                let merge = args
                    .get("merge_strategy")
                    .and_then(Value::as_str)
                    .unwrap_or("concatenate");
                Ok(ToolOutput::user_visible(format_results(&results, merge)))
            }

            other => Err(ZeptoError::Tool(format!(
                "Unknown action '{}'. Valid actions are: run, aggregate",
                other
            ))),
        }
    }
}

/// Merge aggregated sub-agent results using the specified strategy.
///
/// - `"concatenate"` (default) — joins each result as `[Role]: result` separated
///   by blank lines.
/// - `"summarize"` — produces a structured markdown document with `##`/`###`
///   headings. (A real LLM summarization call can be added in a future iteration.)
/// - any other value falls back to `"concatenate"`.
fn format_results(results: &[(String, String)], strategy: &str) -> String {
    match strategy {
        "summarize" => {
            let mut out = String::from("## Aggregated Results\n\n");
            for (role, result) in results {
                out.push_str(&format!("### {}\n{}\n\n", role, result));
            }
            out
        }
        _ => {
            // concatenate (default)
            results
                .iter()
                .map(|(role, result)| format!("[{}]: {}", role, result))
                .collect::<Vec<_>>()
                .join("\n\n")
        }
    }
}

/// Wrapper to convert `Arc<dyn LLMProvider>` into `Box<dyn LLMProvider>`.
///
/// Since `set_provider()` takes `Box<dyn LLMProvider>`, we need this thin wrapper
/// to share the same provider instance via Arc without cloning the provider itself.
struct ProviderRef(Arc<dyn LLMProvider>);

#[async_trait]
impl LLMProvider for ProviderRef {
    fn name(&self) -> &str {
        self.0.name()
    }

    fn default_model(&self) -> &str {
        self.0.default_model()
    }

    async fn chat(
        &self,
        messages: Vec<Message>,
        tools: Vec<ToolDefinition>,
        model: Option<&str>,
        options: ChatOptions,
    ) -> Result<LLMResponse> {
        self.0.chat(messages, tools, model, options).await
    }

    async fn chat_stream(
        &self,
        messages: Vec<Message>,
        tools: Vec<ToolDefinition>,
        model: Option<&str>,
        options: ChatOptions,
    ) -> crate::error::Result<tokio::sync::mpsc::Receiver<crate::providers::StreamEvent>> {
        self.0.chat_stream(messages, tools, model, options).await
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use serde_json::json;

    // Helper to create a DelegateTool for testing
    fn test_delegate_tool(swarm_enabled: bool) -> DelegateTool {
        let mut config = Config::default();
        config.swarm.enabled = swarm_enabled;
        let bus = Arc::new(MessageBus::new());
        let provider: Arc<dyn LLMProvider> =
            Arc::new(crate::providers::claude::ClaudeProvider::new("fake-key"));

        DelegateTool::new(config, provider, bus)
    }

    // -------------------------------------------------------------------------
    // Existing tests (preserved verbatim)
    // -------------------------------------------------------------------------

    #[tokio::test]
    async fn test_delegate_blocked_from_subagent() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("delegate", "sub-123");

        let result = tool
            .execute(json!({"role": "test", "task": "hello"}), &ctx)
            .await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("recursion"),
            "Expected recursion error, got: {}",
            err_msg
        );
    }

    #[tokio::test]
    async fn test_delegate_requires_role() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool.execute(json!({"task": "hello"}), &ctx).await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("role"),
            "Expected role error, got: {}",
            err_msg
        );
    }

    #[tokio::test]
    async fn test_delegate_requires_task() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool.execute(json!({"role": "test"}), &ctx).await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("task"),
            "Expected task error, got: {}",
            err_msg
        );
    }

    #[tokio::test]
    async fn test_delegate_disabled_in_config() {
        let tool = test_delegate_tool(false);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(json!({"role": "test", "task": "hello"}), &ctx)
            .await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("disabled"),
            "Expected disabled error, got: {}",
            err_msg
        );
    }

    #[test]
    fn test_delegate_tool_name() {
        let tool = test_delegate_tool(true);
        assert_eq!(tool.name(), "delegate");
    }

    #[test]
    fn test_delegate_tool_parameters() {
        let tool = test_delegate_tool(true);
        let params = tool.parameters();
        assert_eq!(params["type"], "object");
        assert!(params["properties"]["role"].is_object());
        assert!(params["properties"]["task"].is_object());
        assert!(params["properties"]["tools"].is_object());
    }

    #[test]
    fn test_create_sub_agent_tools_no_whitelist() {
        let tool = test_delegate_tool(true);
        let tools = tool.create_sub_agent_tools(None);
        // Should have basic tools (echo, read, write, list, edit, shell, web_fetch, message)
        // plus memory tools (memory_search, memory_get) since default config enables builtin memory
        assert!(tools.len() >= 8);
        // Should NOT include delegate or spawn
        let names: Vec<&str> = tools.iter().map(|t| t.name()).collect();
        assert!(!names.contains(&"delegate"));
        assert!(!names.contains(&"spawn"));
    }

    #[test]
    fn test_create_sub_agent_tools_with_whitelist() {
        let tool = test_delegate_tool(true);
        let whitelist = vec!["echo".to_string(), "read_file".to_string()];
        let tools = tool.create_sub_agent_tools(Some(&whitelist));
        assert_eq!(tools.len(), 2);
        let names: Vec<&str> = tools.iter().map(|t| t.name()).collect();
        assert!(names.contains(&"echo"));
        assert!(names.contains(&"read_file"));
    }

    // -------------------------------------------------------------------------
    // Task 15: Semaphore tests
    // -------------------------------------------------------------------------

    /// Default SwarmConfig.max_concurrent is 3, so the semaphore should start
    /// with 3 available permits.
    #[test]
    fn test_semaphore_default_capacity() {
        let config = Config::default();
        assert_eq!(config.swarm.max_concurrent, 3);
        let bus = Arc::new(MessageBus::new());
        let provider: Arc<dyn LLMProvider> =
            Arc::new(crate::providers::claude::ClaudeProvider::new("fake-key"));
        let tool = DelegateTool::new(config, provider, bus);
        assert_eq!(tool.semaphore.available_permits(), 3);
    }

    /// Verify that once all permits are taken, `try_acquire` fails (the second
    /// concurrent call would block / be queued).
    #[tokio::test]
    async fn test_semaphore_limits_concurrency() {
        let sem = Arc::new(Semaphore::new(1));
        // Grab the single available permit.
        let _permit = sem.acquire().await.unwrap();
        // A non-blocking attempt to grab another permit must fail.
        assert!(
            sem.try_acquire().is_err(),
            "Semaphore should be exhausted after one permit is held"
        );
    }

    /// A zero max_concurrent value must not produce a zero-capacity semaphore
    /// (which would deadlock every acquire). We clamp it to at least 1.
    #[test]
    fn test_semaphore_zero_max_concurrent_defaults_to_one() {
        let mut config = Config::default();
        config.swarm.max_concurrent = 0;
        let bus = Arc::new(MessageBus::new());
        let provider: Arc<dyn LLMProvider> =
            Arc::new(crate::providers::claude::ClaudeProvider::new("fake-key"));
        let tool = DelegateTool::new(config, provider, bus);
        assert!(
            tool.semaphore.available_permits() >= 1,
            "Zero max_concurrent should clamp to at least 1 permit, got {}",
            tool.semaphore.available_permits()
        );
    }

    // -------------------------------------------------------------------------
    // Task 16: format_results unit tests
    // -------------------------------------------------------------------------

    #[test]
    fn test_format_results_concatenate() {
        let results = vec![
            ("researcher".to_string(), "Found data A".to_string()),
            ("writer".to_string(), "Wrote summary B".to_string()),
        ];
        let out = format_results(&results, "concatenate");
        assert!(
            out.contains("[researcher]: Found data A"),
            "Missing researcher result: {}",
            out
        );
        assert!(
            out.contains("[writer]: Wrote summary B"),
            "Missing writer result: {}",
            out
        );
        // The two blocks must be separated by a blank line.
        assert!(
            out.contains("\n\n"),
            "Expected blank-line separator: {}",
            out
        );
    }

    #[test]
    fn test_format_results_summarize() {
        let results = vec![
            ("analyst".to_string(), "Analysis result".to_string()),
            ("coder".to_string(), "Code review done".to_string()),
        ];
        let out = format_results(&results, "summarize");
        assert!(
            out.starts_with("## Aggregated Results"),
            "Expected h2 header: {}",
            out
        );
        assert!(
            out.contains("### analyst"),
            "Missing analyst header: {}",
            out
        );
        assert!(
            out.contains("Analysis result"),
            "Missing analyst body: {}",
            out
        );
        assert!(out.contains("### coder"), "Missing coder header: {}", out);
        assert!(
            out.contains("Code review done"),
            "Missing coder body: {}",
            out
        );
    }

    #[test]
    fn test_format_results_empty() {
        let results: Vec<(String, String)> = vec![];

        let concat = format_results(&results, "concatenate");
        assert_eq!(concat, "", "Empty concatenate should be empty string");

        let summarize = format_results(&results, "summarize");
        assert!(
            summarize.starts_with("## Aggregated Results"),
            "Empty summarize should still have header: {}",
            summarize
        );
    }

    #[test]
    fn test_format_results_unknown_strategy_falls_back_to_concatenate() {
        let results = vec![("role".to_string(), "result".to_string())];
        let out = format_results(&results, "unknown_strategy");
        assert!(
            out.contains("[role]: result"),
            "Unknown strategy should fall back to concatenate: {}",
            out
        );
    }

    // -------------------------------------------------------------------------
    // Task 16: aggregate action dispatch tests
    // -------------------------------------------------------------------------

    /// aggregate without a 'tasks' key must return an error mentioning "tasks".
    #[tokio::test]
    async fn test_aggregate_requires_tasks() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool.execute(json!({"action": "aggregate"}), &ctx).await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("tasks"),
            "Expected error mentioning 'tasks', got: {}",
            err_msg
        );
    }

    /// A task entry with a role but no 'task' field must error.
    #[tokio::test]
    async fn test_aggregate_task_requires_task_field() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "tasks": [{"role": "analyst"}]
                }),
                &ctx,
            )
            .await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("task"),
            "Expected error mentioning 'task' field, got: {}",
            err_msg
        );
    }

    /// When 'action' is absent the tool should default to "run" and validate
    /// role/task — NOT emit an "unknown action" error.
    #[tokio::test]
    async fn test_action_default_is_run() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        // No 'action' key — should route to the "run" path and fail on missing role.
        let result = tool.execute(json!({"task": "hello"}), &ctx).await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("role"),
            "Missing 'action' should default to 'run' and fail on missing role, got: {}",
            err_msg
        );
        // Must NOT be an "Unknown action" error.
        assert!(
            !err_msg.contains("Unknown action"),
            "Should not get unknown-action error when action is absent: {}",
            err_msg
        );
    }

    // -------------------------------------------------------------------------
    // Task 17: SwarmScratchpad integration test
    // -------------------------------------------------------------------------

    #[test]
    fn test_delegate_has_scratchpad() {
        let tool = test_delegate_tool(true);
        // Scratchpad starts empty
        let rt = tokio::runtime::Runtime::new().unwrap();
        assert!(rt.block_on(tool.scratchpad().is_empty()));
    }

    /// An unrecognised action value must produce an error containing the bad value.
    #[tokio::test]
    async fn test_unknown_action_errors() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(json!({"action": "foo", "role": "r", "task": "t"}), &ctx)
            .await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("foo") || err_msg.contains("Unknown action"),
            "Expected unknown-action error, got: {}",
            err_msg
        );
    }

    /// The JSON schema returned by `parameters()` must expose all new fields
    /// (`action`, `tasks`, `merge_strategy`) alongside the original ones.
    #[test]
    fn test_parameters_include_aggregate_fields() {
        let tool = test_delegate_tool(true);
        let params = tool.parameters();
        let props = &params["properties"];

        // New fields
        assert!(
            props["action"].is_object(),
            "action field missing from schema"
        );
        assert!(
            props["tasks"].is_object(),
            "tasks field missing from schema"
        );
        assert!(
            props["merge_strategy"].is_object(),
            "merge_strategy field missing from schema"
        );

        // Existing fields must still be present
        assert!(props["role"].is_object(), "role field missing from schema");
        assert!(props["task"].is_object(), "task field missing from schema");
        assert!(
            props["tools"].is_object(),
            "tools field missing from schema"
        );
    }

    // -------------------------------------------------------------------------
    // Parallel fan-out tests
    // -------------------------------------------------------------------------

    /// The parameters schema must expose the `parallel` field.
    #[test]
    fn test_parameters_include_parallel_field() {
        let tool = test_delegate_tool(true);
        let params = tool.parameters();
        let props = &params["properties"];
        assert!(
            props["parallel"].is_object(),
            "parallel field missing from schema"
        );
        assert_eq!(
            props["parallel"]["type"], "boolean",
            "parallel should be boolean"
        );
    }

    /// aggregate with parallel=true but missing 'task' field in a spec must
    /// still error before any execution (validation happens upfront).
    #[tokio::test]
    async fn test_parallel_aggregate_validates_upfront() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "parallel": true,
                    "tasks": [
                        {"role": "a", "task": "valid"},
                        {"role": "b"}
                    ]
                }),
                &ctx,
            )
            .await;
        assert!(result.is_err());
        let err_msg = result.unwrap_err().to_string();
        assert!(
            err_msg.contains("task"),
            "Expected upfront validation error for missing 'task', got: {}",
            err_msg
        );
    }

    /// aggregate with parallel=false (default) and missing 'task' in a spec
    /// must also error upfront (same validation path).
    #[tokio::test]
    async fn test_sequential_aggregate_validates_upfront() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "tasks": [{"role": "c"}]
                }),
                &ctx,
            )
            .await;
        assert!(result.is_err());
    }

    /// aggregate with an empty 'tasks' array and parallel=true must succeed
    /// (no sub-agents to run, returns empty result).
    #[tokio::test]
    async fn test_parallel_aggregate_empty_tasks() {
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "parallel": true,
                    "tasks": []
                }),
                &ctx,
            )
            .await;
        // Empty tasks array is valid — returns empty concatenation.
        assert!(result.is_ok());
        let output = result.unwrap();
        assert!(output.for_llm.is_empty());
    }

    /// parallel=false should be the default when the key is absent.
    #[tokio::test]
    async fn test_parallel_defaults_to_false() {
        // This is a structural test — we can't run real sub-agents without
        // an API key, but we CAN verify that the default does not trigger
        // the parallel info log. We verify indirectly by checking that an
        // empty tasks array in default mode produces the same result as
        // parallel=false.
        let tool = test_delegate_tool(true);
        let ctx = ToolContext::new().with_channel("telegram", "chat-1");

        let result_default = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "tasks": []
                }),
                &ctx,
            )
            .await;
        let result_explicit = tool
            .execute(
                json!({
                    "action": "aggregate",
                    "parallel": false,
                    "tasks": []
                }),
                &ctx,
            )
            .await;

        assert!(result_default.is_ok());
        assert!(result_explicit.is_ok());
        assert_eq!(
            result_default.unwrap().for_llm,
            result_explicit.unwrap().for_llm
        );
    }

    /// The description must mention both execution modes.
    #[test]
    fn test_description_mentions_execution_modes() {
        let tool = test_delegate_tool(true);
        let desc = tool.description();
        assert!(
            desc.contains("Parallel") && desc.contains("Sequential"),
            "Description should mention both Parallel and Sequential modes: {}",
            desc
        );
        assert!(
            desc.contains("ask the user"),
            "Description should instruct the agent to ask the user: {}",
            desc
        );
    }
}