merge: integrate oss branch — budget gating, MCP back-channel, events, loopback-only bind

Key features from OSS branch:
- Budget gating: rolling per-provider spend caps with BUDGET_EXCEEDED task state
- Agent MCP back-channel: runners mint tokens; /mcp endpoint resolves them
- Chatbot MCP server at /chatbot/mcp (requires api_token in config)
- Event timeline: unified event stream replacing ad-hoc question/summary flows
- Loopback-only default bind (127.0.0.1:8484); external_bind_allowed=true to expose
- repository_url required on task creation (enforces traceability)
- Auto-checker: spawns verification task after each execution

Conflict resolutions:
- serve.go: keep cfg.Runners.XEnabled() conditional registration from main + agentRegistry from oss
- config.go: keep RunnersConfig from main + BudgetConfig/ExternalBindAllowed from oss
- server.go: handleStaticFiles (base-path rewrite) kept; deduplicated duplicate from both sides
- executor/claude.go: accepted oss deletion (ClaudeRunner replaced by ContainerRunner)
- container_test.go: added noopChannel + AgentChannel parameter to Run call

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2 files changed, 77 insertions, 0 deletions

diff --git a/internal/llm/client.go b/internal/llm/client.go
index 613ebe5..e3a6102 100644
--- a/internal/llm/client.go
+++ b/internal/llm/client.go
@@ -34,6 +34,38 @@ type Client struct {
 type Message struct {
 	Role    string `json:"role"`
 	Content string `json:"content"`
+	// ToolCalls is set on assistant messages that request tool invocations.
+	ToolCalls []ToolCall `json:"tool_calls,omitempty"`
+	// ToolCallID links a role:"tool" result message to the call it answers.
+	ToolCallID string `json:"tool_call_id,omitempty"`
+	// Name carries the tool name on role:"tool" result messages.
+	Name string `json:"name,omitempty"`
+}
+
+// Tool declares a function the model may call (OpenAI function-calling format).
+type Tool struct {
+	Type     string       `json:"type"` // always "function"
+	Function ToolFunction `json:"function"`
+}
+
+// ToolFunction describes a callable function and its JSON-schema parameters.
+type ToolFunction struct {
+	Name        string         `json:"name"`
+	Description string         `json:"description"`
+	Parameters  map[string]any `json:"parameters,omitempty"`
+}
+
+// ToolCall is a function invocation the model emitted in its response.
+type ToolCall struct {
+	ID       string           `json:"id"`
+	Type     string           `json:"type"` // "function"
+	Function ToolCallFunction `json:"function"`
+}
+
+// ToolCallFunction holds the called function's name and raw JSON arguments.
+type ToolCallFunction struct {
+	Name      string `json:"name"`
+	Arguments string `json:"arguments"`
 }
 
 // ChatRequest captures the parameters of a single Chat or ChatStream call.
@@ -45,6 +77,7 @@ type ChatRequest struct {
 	Temperature  *float64
 	MaxTokens    int
 	ResponseJSON bool
+	Tools        []Tool
 }
 
 // ChatResponse is the aggregated result of a chat completion.
@@ -54,6 +87,7 @@ type ChatResponse struct {
 	OutputTokens int
 	Model        string
 	FinishReason string
+	ToolCalls    []ToolCall
 }
 
 // Chat performs a non-streaming chat completion. Rate-limit errors (HTTP 429,
@@ -87,6 +121,7 @@ func (c *Client) Chat(ctx context.Context, req ChatRequest) (*ChatResponse, erro
 			OutputTokens: oai.Usage.CompletionTokens,
 			Model:        oai.Model,
 			FinishReason: oai.Choices[0].FinishReason,
+			ToolCalls:    oai.Choices[0].Message.ToolCalls,
 		}
 		return nil
 	})
@@ -133,6 +168,7 @@ func (c *Client) buildRequestBody(req ChatRequest, stream bool) ([]byte, error)
 		Model:    model,
 		Messages: req.Messages,
 		Stream:   stream,
+		Tools:    req.Tools,
 	}
 	if req.Temperature != nil {
 		payload.Temperature = req.Temperature
@@ -301,6 +337,7 @@ type openAIRequest struct {
 	Stream         bool            `json:"stream,omitempty"`
 	StreamOptions  *streamOptions  `json:"stream_options,omitempty"`
 	ResponseFormat *responseFormat `json:"response_format,omitempty"`
+	Tools          []Tool          `json:"tools,omitempty"`
 }
 
 type streamOptions struct {
diff --git a/internal/llm/client_test.go b/internal/llm/client_test.go
index 8257836..7533a8a 100644
--- a/internal/llm/client_test.go
+++ b/internal/llm/client_test.go
@@ -157,3 +157,43 @@ func TestErrFromStatus_RateLimitMarker(t *testing.T) {
 		t.Errorf("error should embed retry-after, got: %v", err)
 	}
 }
+
+func TestChat_ToolsRoundTrip(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body openAIRequest
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Fatalf("decode body: %v", err)
+		}
+		if len(body.Tools) != 1 || body.Tools[0].Function.Name != "do_thing" {
+			t.Errorf("tools not forwarded in request: %+v", body.Tools)
+		}
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintln(w, `{
+			"model": "m",
+			"choices": [{"message": {"role": "assistant", "content": "",
+				"tool_calls": [{"id": "call_1", "type": "function",
+					"function": {"name": "do_thing", "arguments": "{\"x\":1}"}}]},
+				"finish_reason": "tool_calls"}],
+			"usage": {"prompt_tokens": 3, "completion_tokens": 4}
+		}`)
+	}))
+	defer srv.Close()
+
+	c := &Client{Endpoint: srv.URL + "/v1", Model: "m"}
+	resp, err := c.Chat(context.Background(), ChatRequest{
+		Messages: []Message{{Role: "user", Content: "go"}},
+		Tools: []Tool{{Type: "function", Function: ToolFunction{
+			Name: "do_thing", Description: "d", Parameters: map[string]any{"type": "object"},
+		}}},
+	})
+	if err != nil {
+		t.Fatalf("Chat: %v", err)
+	}
+	if len(resp.ToolCalls) != 1 {
+		t.Fatalf("expected 1 tool call, got %d", len(resp.ToolCalls))
+	}
+	tc := resp.ToolCalls[0]
+	if tc.ID != "call_1" || tc.Function.Name != "do_thing" || tc.Function.Arguments != `{"x":1}` {
+		t.Errorf("unexpected tool call parsed: %+v", tc)
+	}
+}