Your AI agent skills, finally organized. A macOS app to browse, edit, and manage skills across Claude Code, Cursor, Codex, Windsurf, and Amp.
import Foundation
// MARK: - Permission flow
/// One option presented to the user when an agent asks to use a tool.
/// `kind` is a stable string used by the UI to color the button (e.g. `allow_once`,
/// `allow_always`, `reject_once`, `reject_always`).
struct PermissionOption: Codable, Sendable, Identifiable {
let kind: String
let name: String
let optionId: String
var id: String { optionId }
}
/// Optional diff payload attached to a permission ask. When present the UI shows the
/// before/after content side-by-side so the user can review the proposed change before
/// clicking Allow. Required for any tool that mutates disk (Write / Edit / MultiEdit).
struct PermissionDiffPreview: Sendable {
let path: String
let originalText: String?
let proposedText: String
let existedBefore: Bool
}
/// Wraps a parked permission ask so the UI can resume it once the user clicks.
struct PermissionRequest: Identifiable, @unchecked Sendable {
let id: UUID = UUID()
let title: String
let options: [PermissionOption]
let diffPreview: PermissionDiffPreview?
let continuation: CheckedContinuation<PermissionResponse, Error>
}
/// Result of a permission ask. `cancelled == true` means the user dismissed without choosing
/// a specific option; otherwise `optionId` identifies the chosen option.
struct PermissionResponse: Sendable {
let optionId: String?
let cancelled: Bool
static func choice(_ optionId: String) -> PermissionResponse {
PermissionResponse(optionId: optionId, cancelled: false)
}
static var cancelled: PermissionResponse {
PermissionResponse(optionId: nil, cancelled: true)
}
}
// MARK: - Activity feed
/// One step in the agent's chronological activity feed for a turn. The UI shows these
/// to give the user real-time visibility into what the agent is doing — Claude
/// generating long tool inputs can take minutes; without an activity feed it looks like
/// the app is frozen.
struct AgentActivity: Identifiable, Sendable, Equatable {
enum Kind: Sendable, Equatable {
case thinking
case toolCall(name: String)
case toolInputDraft(name: String, charCount: Int)
case toolResult(name: String)
case info
}
enum Status: Sendable, Equatable {
case running
case done // tool ran successfully (read-only / non-mutating)
case applied // mutating tool (Write/Edit/MultiEdit) successfully changed disk
case failed
}
/// Optional rich payload so the UI can show what each step actually did when
/// expanded. For Write/Edit this carries the diff; for Read/Bash the raw input/output.
struct Payload: Sendable, Equatable {
var filePath: String?
var originalText: String?
var proposedText: String?
var existedBefore: Bool?
var rawInput: String? // pretty-printed JSON of tool input
var resultText: String? // tool_result content (truncated)
}
let id: UUID
var kind: Kind
var title: String
var detail: String?
var startedAt: Date
var status: Status
var payload: Payload
init(
id: UUID = UUID(),
kind: Kind,
title: String,
detail: String? = nil,
startedAt: Date = Date(),
status: Status = .running,
payload: Payload = Payload()
) {
self.id = id
self.kind = kind
self.title = title
self.detail = detail
self.startedAt = startedAt
self.status = status
self.payload = payload
}
}
// MARK: - Pending writes
/// A file write proposed by the agent. Surfaced in the chat as a diff for the user to
/// accept or reject. Both Claude (`Write` / `Edit` / `MultiEdit`) and Codex
/// (`item/fileChange/*`) populate this.
struct PendingWrite: Sendable {
let path: String
let content: String
let originalText: String?
let originalData: Data?
let existedBefore: Bool
/// True when the agent itself performed the disk write (the case for direct CLI
/// agents). On reject the UI restores from the captured snapshot.
let agentDidWrite: Bool
init(
path: String,
content: String,
originalText: String?,
originalData: Data?,
existedBefore: Bool,
agentDidWrite: Bool = false
) {
self.path = path
self.content = content
self.originalText = originalText
self.originalData = originalData
self.existedBefore = existedBefore
self.agentDidWrite = agentDidWrite
}
}
// MARK: - Errors
enum AgentError: Error, LocalizedError {
case noSession
case binaryNotInstalled(toolName: String, installURL: URL)
case agentTooOld(toolName: String, found: String, minimum: String)
case launchFailed(String)
case processExitedDuringConnect(String)
case connectTimedOut(stage: String)
var errorDescription: String? {
switch self {
case .noSession:
"No active agent session."
case .binaryNotInstalled(let name, _):
"\(name) isn't installed."
case .agentTooOld(let name, let found, let minimum):
"\(name) v\(found) is too old. Update to v\(minimum) or newer."
case .launchFailed(let detail):
"Failed to launch agent: \(detail)"
case .processExitedDuringConnect(let detail):
"Agent exited before initializing.\n\n\(detail)"
case .connectTimedOut(let stage):
"Connection timed out (\(stage))."
}
}
}
// MARK: - Helpers
enum AgentDataDecoding {
/// Best-effort decode of bytes to a Swift String. Tries UTF-8 then UTF-16.
static func text(from data: Data?) -> String? {
guard let data else { return nil }
return String(data: data, encoding: .utf8)
?? String(data: data, encoding: .utf16)
}
}