chat_context_inject

Add external context like memory recall, documentation, or runtime state to chat sessions before sending messages. Injects content as messages for AI completions.

Instructions

Inject external context into a session before the next message send. Use this to pipe in Aiana memory recall, documentation snippets, or runtime state. The injected content is stored as a message and included in the next completion call.

Input Schema

TableJSON Schema

Name	Required	Description
`sessionId`	Yes	UUID of the session to inject context into.
`context`	Yes	Context content to inject (markdown, JSON, plain text).
`role`	No	Role for the injected context message. Default: system.

Implementation Reference

src/layers/messages.ts:107-120 (handler)

The injectContext function implements the core logic for chat_context_inject tool. It creates a new message in the session with the provided context content and role (system or user), adds metadata tracking the injection timestamp, and returns the created message ID.

export async function injectContext(
  adapter: ChatAdapter,
  sessionId: string,
  context: string,
  role: "system" | "user" = "system",
): Promise<{ messageId: string; sessionId: string; role: string }> {
  const msg = await adapter.addMessage({
    sessionId,
    role,
    content: context,
    metadata: { injected: true, injectedAt: new Date().toISOString() },
  });
  return { messageId: msg.id, sessionId, role };
}

src/app.ts:275-305 (registration)

Registration of the chat_context_inject tool in the tools array. Defines the tool's name, description, input schema (sessionId, context, role), and executes the injectContext function from layers.messages module.

{
  name: "chat_context_inject",
  description:
    "Inject external context into a session before the next message send. Use this to pipe in Aiana memory recall, documentation snippets, or runtime state. The injected content is stored as a message and included in the next completion call.",
  inputSchema: {
    type: "object",
    properties: {
      sessionId: {
        type: "string",
        description: "UUID of the session to inject context into.",
      },
      context: {
        type: "string",
        description: "Context content to inject (markdown, JSON, plain text).",
      },
      role: {
        type: "string",
        enum: ["system", "user"],
        description: "Role for the injected context message. Default: system.",
      },
    },
    required: ["sessionId", "context"],
  },
  execute: async (args) =>
    layers.messages.injectContext(
      adapter,
      args.sessionId as string,
      args.context as string,
      args.role as "system" | "user" | undefined,
    ),
},

src/types.ts:30-40 (schema)

ChatMessage interface defining the data structure for messages, including role field which supports 'system' and 'user' values used by the context injection feature. The metadata field allows storing injection tracking information.

export interface ChatMessage {
  id: string;             // UUID v4
  sessionId: string;
  role: "user" | "assistant" | "system";
  content: string;
  model?: ChatModel;      // set on assistant messages
  inputTokens?: number;   // set on assistant messages
  outputTokens?: number;  // set on assistant messages
  timestamp: string;      // ISO-8601
  metadata?: Record<string, unknown>;
}

@git-fabric/chat