Add tool message types; show tool specific information in webview; separate command from output; add abort button to task card

This commit is contained in:
Saoud Rizwan
2024-07-09 15:05:29 -04:00
parent 771c612d8a
commit d63aef015a
7 changed files with 191 additions and 74 deletions

View File

@@ -9,7 +9,7 @@ import * as path from "path"
import { serializeError } from "serialize-error"
import { DEFAULT_MAX_REQUESTS_PER_TASK } from "./shared/Constants"
import { Tool, ToolName } from "./shared/Tool"
import { ClaudeAsk, ClaudeSay, ExtensionMessage } from "./shared/ExtensionMessage"
import { ClaudeAsk, ClaudeSay, ClaudeSayTool, ExtensionMessage } from "./shared/ExtensionMessage"
import * as vscode from "vscode"
import pWaitFor from "p-wait-for"
import { ClaudeAskResponse } from "./shared/WebviewMessage"
@@ -214,7 +214,7 @@ export class ClaudeDev {
await this.providerRef.deref()?.postStateToWebview()
// Get all relevant context for the task
const filesInCurrentDir = await this.listFiles(".")
const filesInCurrentDir = await this.listFiles(".", false)
// This first message kicks off a task, it is not included in every subsequent message. This is a good place to give all the relevant context to a task, instead of having Claude request for it using tools.
let userPrompt = `# Task
@@ -235,7 +235,7 @@ ${filesInCurrentDir}`
${activeEditorContents}`
}
await this.say("text", userPrompt)
await this.say("text", task)
let totalInputTokens = 0
let totalOutputTokens = 0
@@ -255,10 +255,10 @@ ${activeEditorContents}`
//this.say("task_completed", `Task completed. Total API usage cost: ${totalCost}`)
break
} else {
this.say(
"tool",
"Claude responded with only text blocks but has not called attempt_completion yet. Forcing him to continue with task..."
)
// this.say(
// "tool",
// "Claude responded with only text blocks but has not called attempt_completion yet. Forcing him to continue with task..."
// )
userPrompt =
"Ask yourself if you have completed the user's task. If you have, use the attempt_completion tool, otherwise proceed to the next step. (This is an automated message, so do not respond to it conversationally. Just proceed with the task.)"
}
@@ -305,38 +305,46 @@ ${activeEditorContents}`
const diffResult = diff.createPatch(filePath, originalContent, newContent)
if (diffResult) {
await fs.writeFile(filePath, newContent)
this.say("tool", JSON.stringify({ tool: "editedExistingFile", path: filePath, diff: diffResult } as ClaudeSayTool))
return `Changes applied to ${filePath}:\n${diffResult}`
} else {
this.say("tool", JSON.stringify({ tool: "editedExistingFile", path: filePath } as ClaudeSayTool))
return `Tool succeeded, however there were no changes detected to ${filePath}`
}
} else {
await fs.mkdir(path.dirname(filePath), { recursive: true })
await fs.writeFile(filePath, newContent)
this.say("tool", JSON.stringify({ tool: "newFileCreated", path: filePath, content: newContent } as ClaudeSayTool))
return `New file created and content written to ${filePath}`
}
} catch (error) {
const errorString = `Error writing file: ${JSON.stringify(serializeError(error))}`
this.say("error", errorString)
this.say("error", JSON.stringify(serializeError(error)))
return errorString
}
}
async readFile(filePath: string): Promise<string> {
try {
return await fs.readFile(filePath, "utf-8")
const content = await fs.readFile(filePath, "utf-8")
this.say("tool", JSON.stringify({ tool: "readFile", path: filePath } as ClaudeSayTool))
return content
} catch (error) {
const errorString = `Error reading file: ${JSON.stringify(serializeError(error))}`
this.say("error", errorString)
this.say("error", JSON.stringify(serializeError(error)))
return errorString
}
}
async listFiles(dirPath: string): Promise<string> {
async listFiles(dirPath: string, shouldLog: boolean = true): Promise<string> {
// If the extension is run without a workspace open, we are in the root directory and don't want to list all files since it would prompt for permission to access everything
const cwd = process.cwd()
const root = process.platform === "win32" ? path.parse(cwd).root : "/"
const isRoot = cwd === root
if (isRoot) {
if (shouldLog) {
this.say("tool", JSON.stringify({ tool: "listFiles", path: dirPath } as ClaudeSayTool))
}
return "Currently in the root directory. Cannot list all files."
}
@@ -348,19 +356,19 @@ ${activeEditorContents}`
}
// * globs all files in one dir, ** globs files in nested directories
const entries = await glob("*", options)
if (shouldLog) {
this.say("tool", JSON.stringify({ tool: "listFiles", path: dirPath } as ClaudeSayTool))
}
return entries.slice(0, 500).join("\n") // truncate to 500 entries
} catch (error) {
const errorString = `Error listing files and directories: ${JSON.stringify(serializeError(error))}`
this.say("error", errorString)
this.say("error", JSON.stringify(serializeError(error)))
return errorString
}
}
async executeCommand(command: string): Promise<string> {
const { response } = await this.ask(
"command",
`Claude wants to execute the following command:\n${command}\nDo you approve?`
)
const { response } = await this.ask("command", command)
if (response !== "yesButtonTapped") {
return "Command execution was not approved by the user."
}
@@ -378,7 +386,7 @@ ${activeEditorContents}`
const error = e as any
let errorMessage = error.message || JSON.stringify(serializeError(error))
const errorString = `Error executing command:\n${errorMessage}`
this.say("error", errorString)
this.say("error", errorMessage)
return errorString
}
}
@@ -414,7 +422,7 @@ ${activeEditorContents}`
if (this.requestCount >= this.maxRequestsPerTask) {
const { response } = await this.ask(
"request_limit_reached",
`\nClaude has exceeded ${this.maxRequestsPerTask} requests for this task! Would you like to reset the count and proceed?:`
`Claude Dev has reached the maximum number of requests for this task. Would you like to reset the count and allow him to proceed?`
)
if (response === "yesButtonTapped") {
@@ -434,7 +442,7 @@ ${activeEditorContents}`
}
try {
await this.say("api_req_started", JSON.stringify(userContent))
await this.say("api_req_started", JSON.stringify({ request: userContent }))
const response = await this.client.messages.create({
model: "claude-3-5-sonnet-20240620", // https://docs.anthropic.com/en/docs/about-claude/models
max_tokens: 4096,
@@ -448,7 +456,7 @@ ${activeEditorContents}`
let assistantResponses: Anthropic.Messages.ContentBlock[] = []
let inputTokens = response.usage.input_tokens
let outputTokens = response.usage.output_tokens
await this.say("api_req_finished", this.calculateApiCost(inputTokens, outputTokens).toString())
await this.say("api_req_finished", JSON.stringify({ tokensIn: inputTokens, tokensOut: outputTokens, cost: this.calculateApiCost(inputTokens, outputTokens) }))
// A response always returns text content blocks (it's just that before we were iterating over the completion_attempt response before we could append text response, resulting in bug)
for (const contentBlock of response.content) {
@@ -470,10 +478,10 @@ ${activeEditorContents}`
attemptCompletionBlock = contentBlock
} else {
const result = await this.executeTool(toolName, toolInput)
this.say(
"tool",
`\nTool Used: ${toolName}\nTool Input: ${JSON.stringify(toolInput)}\nTool Result: ${result}`
)
// this.say(
// "tool",
// `\nTool Used: ${toolName}\nTool Input: ${JSON.stringify(toolInput)}\nTool Result: ${result}`
// )
toolResults.push({ type: "tool_result", tool_use_id: toolUseId, content: result })
}
}
@@ -483,7 +491,7 @@ ${activeEditorContents}`
this.conversationHistory.push({ role: "assistant", content: assistantResponses })
} else {
// this should never happen! it there's no assistant_responses, that means we got no text or tool_use content blocks from API which we should assume is an error
this.say("error", "Error: No assistant responses found in API response!")
this.say("error", "Unexpected Error: No assistant messages were found in the API response")
this.conversationHistory.push({
role: "assistant",
content: [{ type: "text", text: "Failure: I did not have a response to provide." }],
@@ -499,12 +507,12 @@ ${activeEditorContents}`
attemptCompletionBlock.name as ToolName,
attemptCompletionBlock.input
)
this.say(
"tool",
`\nattempt_completion Tool Used: ${attemptCompletionBlock.name}\nTool Input: ${JSON.stringify(
attemptCompletionBlock.input
)}\nTool Result: ${result}`
)
// this.say(
// "tool",
// `\nattempt_completion Tool Used: ${attemptCompletionBlock.name}\nTool Input: ${JSON.stringify(
// attemptCompletionBlock.input
// )}\nTool Result: ${result}`
// )
if (result === "") {
didCompleteTask = true
result = "The user is satisfied with the result."
@@ -539,7 +547,7 @@ ${activeEditorContents}`
return { didCompleteTask, inputTokens, outputTokens }
} catch (error) {
// only called if the API request fails (executeTool errors are returned back to claude)
this.say("error", `Error calling Claude API: ${JSON.stringify(serializeError(error))}`)
this.say("error", JSON.stringify(serializeError(error)))
return { didCompleteTask: true, inputTokens: 0, outputTokens: 0 }
}
}

View File

@@ -17,4 +17,11 @@ export interface ClaudeMessage {
}
export type ClaudeAsk = "request_limit_reached" | "followup" | "command" | "completion_result"
export type ClaudeSay = "task" | "error" | "api_req_started" | "api_req_finished" | "text" | "tool" | "command_output" | "completion_result"
export type ClaudeSay = "task" | "error" | "api_req_started" | "api_req_finished" | "text" | "tool" | "command_output" | "completion_result"
export interface ClaudeSayTool {
tool: "editedExistingFile" | "newFileCreated" | "readFile" | "listFiles"
path?: string
diff?: string
content?: string
}

View File

@@ -1,5 +1,5 @@
export interface WebviewMessage {
type: "apiKey" | "maxRequestsPerTask" | "webviewDidLaunch" | "newTask" | "askResponse"
type: "apiKey" | "maxRequestsPerTask" | "webviewDidLaunch" | "newTask" | "askResponse" | "abortTask"
text?: string
askResponse?: ClaudeAskResponse
}