fix(tasktool): use agent session key, persist plan on create, add docs

- Add WithToolSessionKey/ToolSessionKey context helpers to pkg/tools/base.go
- Inject opts.SessionKey into ctx in runAgentLoop so tasktool uses the
  same agent-scoped session key as conversation history instead of the
  bare channel:chatID composite (fixes plan cross-agent/scope collision)
- Call Save inside CreatePlan so plans persist after restart even on
  channels that never return a message ID (fixes data loss on non-Telegram)
- Add tasktool section to config/config.example.json and
  docs/tools_configuration.md (P3 documentation gap)

config/config.example.json

@@ -431,6 +431,15 @@
     },
     "write_file": {
       "enabled": true
+    },
+    "tasktool": {
+      "enabled": true,
+      "icons": {
+        "pending": "🔘",
+        "in_progress": "🟡",
+        "completed": "🟢",
+        "failed": "🔴"
+      }
     }
   },
   "heartbeat": {

docs/tools_configuration.md

@@ -206,6 +206,48 @@ The skills tool configures skill discovery and installation via registries like
 }
 ```
 
+## Task Tool
+
+The task tool lets the agent create and track a step-by-step execution plan visible to the user in real time.
+
+### Config
+
+| Config              | Type   | Default | Description                            |
+| ------------------- | ------ | ------- | -------------------------------------- |
+| `enabled`           | bool   | true    | Enable the task planning tool          |
+| `icons.pending`     | string | 🔘      | Icon shown for pending tasks           |
+| `icons.in_progress` | string | 🟡      | Icon shown for tasks in progress       |
+| `icons.completed`   | string | 🟢      | Icon shown for completed tasks         |
+| `icons.failed`      | string | 🔴      | Icon shown for failed tasks            |
+
+### Configuration Example
+
+```json
+{
+  "tools": {
+    "tasktool": {
+      "enabled": true,
+      "icons": {
+        "pending": "🔘",
+        "in_progress": "🟡",
+        "completed": "🟢",
+        "failed": "🔴"
+      }
+    }
+  }
+}
+```
+
+### Environment Variables
+
+| Variable                                     | Description              |
+| -------------------------------------------- | ------------------------ |
+| `PICOCLAW_TOOLS_TASK_TOOL_ENABLED`           | Enable/disable the tool  |
+| `PICOCLAW_TOOLS_TASK_TOOL_ICONS_PENDING`     | Pending task icon        |
+| `PICOCLAW_TOOLS_TASK_TOOL_ICONS_IN_PROGRESS` | In-progress task icon    |
+| `PICOCLAW_TOOLS_TASK_TOOL_ICONS_COMPLETED`   | Completed task icon      |
+| `PICOCLAW_TOOLS_TASK_TOOL_ICONS_FAILED`      | Failed task icon         |
+
 ## Environment Variables
 
 All configuration options can be overridden via environment variables with the format `PICOCLAW_TOOLS_<SECTION>_<KEY>`:

pkg/agent/loop.go

@@ -798,6 +798,8 @@ func (al *AgentLoop) runAgentLoop(
 	agent.Sessions.AddMessage(opts.SessionKey, "user", opts.UserMessage)
 
 	// 3. Run LLM iteration loop
+	// Inject session key so tools (e.g. tasktool) can look it up from context.
+	ctx = tools.WithToolSessionKey(ctx, opts.SessionKey)
 	finalContent, iteration, err := al.runLLMIteration(ctx, agent, messages, opts)
 	if err != nil {
 		return "", err

pkg/session/tasks.go

@@ -97,6 +97,9 @@ func (tm *TaskManager) CreatePlan(sessionKey string, tasks []Task) *SessionTasks
 	}
 	copy(st.Tasks, tasks)
 	tm.tasks[sessionKey] = st
+	// Persist immediately so the plan survives a restart even when no
+	// message ID is available (e.g. channels that don't return a message ID).
+	go func() { _ = tm.Save(sessionKey) }()
 	return st
 }
 

pkg/tools/base.go

@@ -21,8 +21,9 @@ type Tool interface {
 type toolCtxKey struct{ name string }
 
 var (
-	ctxKeyChannel = &toolCtxKey{"channel"}
-	ctxKeyChatID  = &toolCtxKey{"chatID"}
+	ctxKeyChannel    = &toolCtxKey{"channel"}
+	ctxKeyChatID     = &toolCtxKey{"chatID"}
+	ctxKeySessionKey = &toolCtxKey{"sessionKey"}
 )
 
 // WithToolContext returns a child context carrying channel and chatID.
@@ -32,6 +33,11 @@ func WithToolContext(ctx context.Context, channel, chatID string) context.Contex
 	return ctx
 }
 
+// WithToolSessionKey returns a child context carrying the session key.
+func WithToolSessionKey(ctx context.Context, sessionKey string) context.Context {
+	return context.WithValue(ctx, ctxKeySessionKey, sessionKey)
+}
+
 // ToolChannel extracts the channel from ctx, or "" if unset.
 func ToolChannel(ctx context.Context) string {
 	v, _ := ctx.Value(ctxKeyChannel).(string)
@@ -44,6 +50,12 @@ func ToolChatID(ctx context.Context) string {
 	return v
 }
 
+// ToolSessionKey extracts the session key from ctx, or "" if unset.
+func ToolSessionKey(ctx context.Context) string {
+	v, _ := ctx.Value(ctxKeySessionKey).(string)
+	return v
+}
+
 // AsyncCallback is a function type that async tools use to notify completion.
 // When an async tool finishes its work, it calls this callback with the result.
 //

pkg/tools/tasktool.go

@@ -98,11 +98,14 @@ func (t *TaskTool) Execute(ctx context.Context, args map[string]any) *ToolResult
 
 	channel := ToolChannel(ctx)
 	chatID := ToolChatID(ctx)
-	// We use the same combination for task state as session manager might.
-	// But note: AgentLoop uses scopes out of routes. We'll use channel:chatID as implicit for now
-	// To be perfectly aligned with SessionKey, we'd need to extract SessionKey from context.
-	// We'll add SessionKey to context later if needed, or just use channel:chatID for tasks since planning is chat-specific.
-	sessionKey := fmt.Sprintf("%s:%s", channel, chatID)
+	// Use the agent-scoped session key injected by the agent loop so that task
+	// plans are stored under the same key used for conversation history.
+	// Fall back to channel:chatID only when running outside of an agent loop
+	// (e.g. unit tests or direct tool invocations).
+	sessionKey := ToolSessionKey(ctx)
+	if sessionKey == "" {
+		sessionKey = fmt.Sprintf("%s:%s", channel, chatID)
+	}
 
 	action, ok := args["action"].(string)
 	if !ok {