ag-cortex 0.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (162) hide show
  1. package/.agent/commands/test-browser.md +339 -0
  2. package/.agent/rules/00-constitution.md +46 -0
  3. package/.agent/rules/project-rules.md +49 -0
  4. package/.agent/skills/agent-browser/SKILL.md +223 -0
  5. package/.agent/skills/agent-native-architecture/SKILL.md +435 -0
  6. package/.agent/skills/agent-native-architecture/references/action-parity-discipline.md +409 -0
  7. package/.agent/skills/agent-native-architecture/references/agent-execution-patterns.md +467 -0
  8. package/.agent/skills/agent-native-architecture/references/agent-native-testing.md +582 -0
  9. package/.agent/skills/agent-native-architecture/references/architecture-patterns.md +478 -0
  10. package/.agent/skills/agent-native-architecture/references/dynamic-context-injection.md +338 -0
  11. package/.agent/skills/agent-native-architecture/references/files-universal-interface.md +301 -0
  12. package/.agent/skills/agent-native-architecture/references/from-primitives-to-domain-tools.md +359 -0
  13. package/.agent/skills/agent-native-architecture/references/mcp-tool-design.md +506 -0
  14. package/.agent/skills/agent-native-architecture/references/mobile-patterns.md +871 -0
  15. package/.agent/skills/agent-native-architecture/references/product-implications.md +443 -0
  16. package/.agent/skills/agent-native-architecture/references/refactoring-to-prompt-native.md +317 -0
  17. package/.agent/skills/agent-native-architecture/references/self-modification.md +269 -0
  18. package/.agent/skills/agent-native-architecture/references/shared-workspace-architecture.md +680 -0
  19. package/.agent/skills/agent-native-architecture/references/system-prompt-design.md +250 -0
  20. package/.agent/skills/agent-native-reviewer/SKILL.md +246 -0
  21. package/.agent/skills/andrew-kane-gem-writer/SKILL.md +184 -0
  22. package/.agent/skills/andrew-kane-gem-writer/references/database-adapters.md +231 -0
  23. package/.agent/skills/andrew-kane-gem-writer/references/module-organization.md +121 -0
  24. package/.agent/skills/andrew-kane-gem-writer/references/rails-integration.md +183 -0
  25. package/.agent/skills/andrew-kane-gem-writer/references/resources.md +119 -0
  26. package/.agent/skills/andrew-kane-gem-writer/references/testing-patterns.md +261 -0
  27. package/.agent/skills/ankane-readme-writer/SKILL.md +50 -0
  28. package/.agent/skills/architecture-strategist/SKILL.md +52 -0
  29. package/.agent/skills/best-practices-researcher/SKILL.md +100 -0
  30. package/.agent/skills/bug-reproduction-validator/SKILL.md +67 -0
  31. package/.agent/skills/code-simplicity-reviewer/SKILL.md +85 -0
  32. package/.agent/skills/coding-tutor/.claude-plugin/plugin.json +9 -0
  33. package/.agent/skills/coding-tutor/README.md +37 -0
  34. package/.agent/skills/coding-tutor/commands/quiz-me.md +1 -0
  35. package/.agent/skills/coding-tutor/commands/sync-tutorials.md +25 -0
  36. package/.agent/skills/coding-tutor/commands/teach-me.md +1 -0
  37. package/.agent/skills/coding-tutor/skills/coding-tutor/SKILL.md +214 -0
  38. package/.agent/skills/coding-tutor/skills/coding-tutor/scripts/create_tutorial.py +202 -0
  39. package/.agent/skills/coding-tutor/skills/coding-tutor/scripts/index_tutorials.py +203 -0
  40. package/.agent/skills/coding-tutor/skills/coding-tutor/scripts/quiz_priority.py +190 -0
  41. package/.agent/skills/coding-tutor/skills/coding-tutor/scripts/setup_tutorials.py +132 -0
  42. package/.agent/skills/compound-docs/SKILL.md +510 -0
  43. package/.agent/skills/compound-docs/assets/critical-pattern-template.md +34 -0
  44. package/.agent/skills/compound-docs/assets/resolution-template.md +93 -0
  45. package/.agent/skills/compound-docs/references/yaml-schema.md +65 -0
  46. package/.agent/skills/compound-docs/schema.yaml +176 -0
  47. package/.agent/skills/create-agent-skills/SKILL.md +299 -0
  48. package/.agent/skills/create-agent-skills/references/api-security.md +226 -0
  49. package/.agent/skills/create-agent-skills/references/be-clear-and-direct.md +531 -0
  50. package/.agent/skills/create-agent-skills/references/best-practices.md +404 -0
  51. package/.agent/skills/create-agent-skills/references/common-patterns.md +595 -0
  52. package/.agent/skills/create-agent-skills/references/core-principles.md +437 -0
  53. package/.agent/skills/create-agent-skills/references/executable-code.md +175 -0
  54. package/.agent/skills/create-agent-skills/references/iteration-and-testing.md +474 -0
  55. package/.agent/skills/create-agent-skills/references/official-spec.md +185 -0
  56. package/.agent/skills/create-agent-skills/references/recommended-structure.md +168 -0
  57. package/.agent/skills/create-agent-skills/references/skill-structure.md +372 -0
  58. package/.agent/skills/create-agent-skills/references/using-scripts.md +113 -0
  59. package/.agent/skills/create-agent-skills/references/using-templates.md +112 -0
  60. package/.agent/skills/create-agent-skills/references/workflows-and-validation.md +510 -0
  61. package/.agent/skills/create-agent-skills/templates/router-skill.md +73 -0
  62. package/.agent/skills/create-agent-skills/templates/simple-skill.md +33 -0
  63. package/.agent/skills/create-agent-skills/workflows/add-reference.md +96 -0
  64. package/.agent/skills/create-agent-skills/workflows/add-script.md +93 -0
  65. package/.agent/skills/create-agent-skills/workflows/add-template.md +74 -0
  66. package/.agent/skills/create-agent-skills/workflows/add-workflow.md +120 -0
  67. package/.agent/skills/create-agent-skills/workflows/audit-skill.md +138 -0
  68. package/.agent/skills/create-agent-skills/workflows/create-domain-expertise-skill.md +605 -0
  69. package/.agent/skills/create-agent-skills/workflows/create-new-skill.md +191 -0
  70. package/.agent/skills/create-agent-skills/workflows/get-guidance.md +121 -0
  71. package/.agent/skills/create-agent-skills/workflows/upgrade-to-router.md +161 -0
  72. package/.agent/skills/create-agent-skills/workflows/verify-skill.md +204 -0
  73. package/.agent/skills/data-integrity-guardian/SKILL.md +70 -0
  74. package/.agent/skills/data-migration-expert/SKILL.md +97 -0
  75. package/.agent/skills/deployment-verification-agent/SKILL.md +159 -0
  76. package/.agent/skills/design-implementation-reviewer/SKILL.md +85 -0
  77. package/.agent/skills/design-iterator/SKILL.md +197 -0
  78. package/.agent/skills/dhh-rails-reviewer/SKILL.md +45 -0
  79. package/.agent/skills/dhh-rails-style/SKILL.md +184 -0
  80. package/.agent/skills/dhh-rails-style/references/architecture.md +653 -0
  81. package/.agent/skills/dhh-rails-style/references/controllers.md +303 -0
  82. package/.agent/skills/dhh-rails-style/references/frontend.md +510 -0
  83. package/.agent/skills/dhh-rails-style/references/gems.md +266 -0
  84. package/.agent/skills/dhh-rails-style/references/models.md +359 -0
  85. package/.agent/skills/dhh-rails-style/references/testing.md +338 -0
  86. package/.agent/skills/dspy-ruby/SKILL.md +594 -0
  87. package/.agent/skills/dspy-ruby/assets/config-template.rb +359 -0
  88. package/.agent/skills/dspy-ruby/assets/module-template.rb +326 -0
  89. package/.agent/skills/dspy-ruby/assets/signature-template.rb +143 -0
  90. package/.agent/skills/dspy-ruby/references/core-concepts.md +265 -0
  91. package/.agent/skills/dspy-ruby/references/optimization.md +623 -0
  92. package/.agent/skills/dspy-ruby/references/providers.md +305 -0
  93. package/.agent/skills/every-style-editor/SKILL.md +134 -0
  94. package/.agent/skills/every-style-editor/references/EVERY_WRITE_STYLE.md +529 -0
  95. package/.agent/skills/figma-design-sync/SKILL.md +166 -0
  96. package/.agent/skills/file-todos/SKILL.md +251 -0
  97. package/.agent/skills/file-todos/assets/todo-template.md +155 -0
  98. package/.agent/skills/framework-docs-researcher/SKILL.md +83 -0
  99. package/.agent/skills/frontend-design/SKILL.md +42 -0
  100. package/.agent/skills/gemini-imagegen/SKILL.md +237 -0
  101. package/.agent/skills/gemini-imagegen/requirements.txt +2 -0
  102. package/.agent/skills/gemini-imagegen/scripts/compose_images.py +168 -0
  103. package/.agent/skills/gemini-imagegen/scripts/edit_image.py +157 -0
  104. package/.agent/skills/gemini-imagegen/scripts/gemini_images.py +265 -0
  105. package/.agent/skills/gemini-imagegen/scripts/generate_image.py +147 -0
  106. package/.agent/skills/gemini-imagegen/scripts/multi_turn_chat.py +215 -0
  107. package/.agent/skills/git-history-analyzer/SKILL.md +42 -0
  108. package/.agent/skills/git-worktree/SKILL.md +302 -0
  109. package/.agent/skills/git-worktree/scripts/worktree-manager.sh +345 -0
  110. package/.agent/skills/julik-frontend-races-reviewer/SKILL.md +222 -0
  111. package/.agent/skills/kieran-python-reviewer/SKILL.md +104 -0
  112. package/.agent/skills/kieran-rails-reviewer/SKILL.md +86 -0
  113. package/.agent/skills/kieran-typescript-reviewer/SKILL.md +95 -0
  114. package/.agent/skills/lint/SKILL.md +16 -0
  115. package/.agent/skills/pattern-recognition-specialist/SKILL.md +57 -0
  116. package/.agent/skills/performance-oracle/SKILL.md +110 -0
  117. package/.agent/skills/pr-comment-resolver/SKILL.md +69 -0
  118. package/.agent/skills/rclone/SKILL.md +150 -0
  119. package/.agent/skills/rclone/scripts/check_setup.sh +60 -0
  120. package/.agent/skills/repo-research-analyst/SKILL.md +113 -0
  121. package/.agent/skills/security-sentinel/SKILL.md +93 -0
  122. package/.agent/skills/skill-creator/SKILL.md +209 -0
  123. package/.agent/skills/skill-creator/scripts/init_skill.py +304 -0
  124. package/.agent/skills/skill-creator/scripts/package_skill.py +112 -0
  125. package/.agent/skills/skill-creator/scripts/quick_validate.py +72 -0
  126. package/.agent/skills/spec-flow-analyzer/SKILL.md +113 -0
  127. package/.agent/skills/test-agent/SKILL.md +4 -0
  128. package/.agent/workflows/agent-native-audit.md +277 -0
  129. package/.agent/workflows/ask-user-question.md +21 -0
  130. package/.agent/workflows/changelog.md +137 -0
  131. package/.agent/workflows/compound.md +202 -0
  132. package/.agent/workflows/create-agent-skill.md +8 -0
  133. package/.agent/workflows/deepen-plan-research.md +334 -0
  134. package/.agent/workflows/deepen-plan-synthesis.md +182 -0
  135. package/.agent/workflows/deepen-plan.md +79 -0
  136. package/.agent/workflows/feature-video.md +342 -0
  137. package/.agent/workflows/generate-command.md +162 -0
  138. package/.agent/workflows/heal-skill.md +142 -0
  139. package/.agent/workflows/lfg.md +20 -0
  140. package/.agent/workflows/plan-analysis.md +67 -0
  141. package/.agent/workflows/plan-next-steps.md +63 -0
  142. package/.agent/workflows/plan-review.md +33 -0
  143. package/.agent/workflows/plan-synthesis.md +106 -0
  144. package/.agent/workflows/plan.md +49 -0
  145. package/.agent/workflows/report-bug.md +150 -0
  146. package/.agent/workflows/reproduce-bug.md +99 -0
  147. package/.agent/workflows/resolve-parallel.md +34 -0
  148. package/.agent/workflows/resolve-pr-parallel.md +49 -0
  149. package/.agent/workflows/resolve-todo-parallel.md +35 -0
  150. package/.agent/workflows/review-analysis.md +145 -0
  151. package/.agent/workflows/review-synthesis.md +262 -0
  152. package/.agent/workflows/review.md +64 -0
  153. package/.agent/workflows/ship.md +90 -0
  154. package/.agent/workflows/test-command.md +3 -0
  155. package/.agent/workflows/triage.md +310 -0
  156. package/.agent/workflows/work.md +157 -0
  157. package/.agent/workflows/xcode-test.md +332 -0
  158. package/LICENSE +22 -0
  159. package/README.md +49 -0
  160. package/bin/ag-cortex.js +54 -0
  161. package/lib/core.js +165 -0
  162. package/package.json +31 -0
@@ -0,0 +1,871 @@
1
+ <overview>
2
+ Mobile is a first-class platform for agent-native apps. It has unique constraints and opportunities. This guide covers why mobile matters, iOS storage architecture, checkpoint/resume patterns, and cost-aware design.
3
+ </overview>
4
+
5
+ <why_mobile>
6
+ ## Why Mobile Matters
7
+
8
+ Mobile devices offer unique advantages for agent-native apps:
9
+
10
+ ### A File System
11
+ Agents can work with files naturally, using the same primitives that work everywhere else. The filesystem is the universal interface.
12
+
13
+ ### Rich Context
14
+ A walled garden you get access to. Health data, location, photos, calendars—context that doesn't exist on desktop or web. This enables deeply personalized agent experiences.
15
+
16
+ ### Local Apps
17
+ Everyone has their own copy of the app. This opens opportunities that aren't fully realized yet: apps that modify themselves, fork themselves, evolve per-user. App Store policies constrain some of this today, but the foundation is there.
18
+
19
+ ### Cross-Device Sync
20
+ If you use the file system with iCloud, all devices share the same file system. The agent's work on one device appears on all devices—without you having to build a server.
21
+
22
+ ### The Challenge
23
+
24
+ **Agents are long-running. Mobile apps are not.**
25
+
26
+ An agent might need 30 seconds, 5 minutes, or an hour to complete a task. But iOS will background your app after seconds of inactivity, and may kill it entirely to reclaim memory. The user might switch apps, take a call, or lock their phone mid-task.
27
+
28
+ This means mobile agent apps need:
29
+ - **Checkpointing** — Saving state so work isn't lost
30
+ - **Resuming** — Picking up where you left off after interruption
31
+ - **Background execution** — Using the limited time iOS gives you wisely
32
+ - **On-device vs. cloud decisions** — What runs locally vs. what needs a server
33
+ </why_mobile>
34
+
35
+ <ios_storage>
36
+ ## iOS Storage Architecture
37
+
38
+ > **Needs validation:** This is an approach that works well, but better solutions may exist.
39
+
40
+ For agent-native iOS apps, use iCloud Drive's Documents folder for your shared workspace. This gives you **free, automatic multi-device sync** without building a sync layer or running a server.
41
+
42
+ ### Why iCloud Documents?
43
+
44
+ | Approach | Cost | Complexity | Offline | Multi-Device |
45
+ |----------|------|------------|---------|--------------|
46
+ | Custom backend + sync | $$$ | High | Manual | Yes |
47
+ | CloudKit database | Free tier limits | Medium | Manual | Yes |
48
+ | **iCloud Documents** | Free (user's storage) | Low | Automatic | Automatic |
49
+
50
+ iCloud Documents:
51
+ - Uses user's existing iCloud storage (free 5GB, most users have more)
52
+ - Automatic sync across all user's devices
53
+ - Works offline, syncs when online
54
+ - Files visible in Files.app for transparency
55
+ - No server costs, no sync code to maintain
56
+
57
+ ### Implementation: iCloud-First with Local Fallback
58
+
59
+ ```swift
60
+ // Get the iCloud Documents container
61
+ func iCloudDocumentsURL() -> URL? {
62
+ FileManager.default.url(forUbiquityContainerIdentifier: nil)?
63
+ .appendingPathComponent("Documents")
64
+ }
65
+
66
+ // Your shared workspace lives in iCloud
67
+ class SharedWorkspace {
68
+ let rootURL: URL
69
+
70
+ init() {
71
+ // Use iCloud if available, fall back to local
72
+ if let iCloudURL = iCloudDocumentsURL() {
73
+ self.rootURL = iCloudURL
74
+ } else {
75
+ // Fallback to local Documents (user not signed into iCloud)
76
+ self.rootURL = FileManager.default.urls(
77
+ for: .documentDirectory,
78
+ in: .userDomainMask
79
+ ).first!
80
+ }
81
+ }
82
+
83
+ // All file operations go through this root
84
+ func researchPath(for bookId: String) -> URL {
85
+ rootURL.appendingPathComponent("Research/\(bookId)")
86
+ }
87
+
88
+ func journalPath() -> URL {
89
+ rootURL.appendingPathComponent("Journal")
90
+ }
91
+ }
92
+ ```
93
+
94
+ ### Directory Structure in iCloud
95
+
96
+ ```
97
+ iCloud Drive/
98
+ └── YourApp/ # Your app's container
99
+ └── Documents/ # Visible in Files.app
100
+ ├── Journal/
101
+ │ ├── user/
102
+ │ │ └── 2025-01-15.md # Syncs across devices
103
+ │ └── agent/
104
+ │ └── 2025-01-15.md # Agent observations sync too
105
+ ├── Research/
106
+ │ └── {bookId}/
107
+ │ ├── full_text.txt
108
+ │ └── sources/
109
+ ├── Chats/
110
+ │ └── {conversationId}.json
111
+ └── context.md # Agent's accumulated knowledge
112
+ ```
113
+
114
+ ### Handling iCloud File States
115
+
116
+ iCloud files may not be downloaded locally. Handle this:
117
+
118
+ ```swift
119
+ func readFile(at url: URL) throws -> String {
120
+ // iCloud may create .icloud placeholder files
121
+ if url.pathExtension == "icloud" {
122
+ // Trigger download
123
+ try FileManager.default.startDownloadingUbiquitousItem(at: url)
124
+ throw FileNotYetAvailableError()
125
+ }
126
+
127
+ return try String(contentsOf: url, encoding: .utf8)
128
+ }
129
+
130
+ // For writes, use coordinated file access
131
+ func writeFile(_ content: String, to url: URL) throws {
132
+ let coordinator = NSFileCoordinator()
133
+ var error: NSError?
134
+
135
+ coordinator.coordinate(
136
+ writingItemAt: url,
137
+ options: .forReplacing,
138
+ error: &error
139
+ ) { newURL in
140
+ try? content.write(to: newURL, atomically: true, encoding: .utf8)
141
+ }
142
+
143
+ if let error = error { throw error }
144
+ }
145
+ ```
146
+
147
+ ### What iCloud Enables
148
+
149
+ 1. **User starts experiment on iPhone** → Agent creates config file
150
+ 2. **User opens app on iPad** → Same experiment visible, no sync code needed
151
+ 3. **Agent logs observation on iPhone** → Syncs to iPad automatically
152
+ 4. **User edits journal on iPad** → iPhone sees the edit
153
+
154
+ ### Entitlements Required
155
+
156
+ Add to your app's entitlements:
157
+
158
+ ```xml
159
+ <key>com.apple.developer.icloud-container-identifiers</key>
160
+ <array>
161
+ <string>iCloud.com.yourcompany.yourapp</string>
162
+ </array>
163
+ <key>com.apple.developer.icloud-services</key>
164
+ <array>
165
+ <string>CloudDocuments</string>
166
+ </array>
167
+ <key>com.apple.developer.ubiquity-container-identifiers</key>
168
+ <array>
169
+ <string>iCloud.com.yourcompany.yourapp</string>
170
+ </array>
171
+ ```
172
+
173
+ ### When NOT to Use iCloud Documents
174
+
175
+ - **Sensitive data** - Use Keychain or encrypted local storage instead
176
+ - **High-frequency writes** - iCloud sync has latency; use local + periodic sync
177
+ - **Large media files** - Consider CloudKit Assets or on-demand resources
178
+ - **Shared between users** - iCloud Documents is single-user; use CloudKit for sharing
179
+ </ios_storage>
180
+
181
+ <background_execution>
182
+ ## Background Execution & Resumption
183
+
184
+ > **Needs validation:** These patterns work but better solutions may exist.
185
+
186
+ Mobile apps can be suspended or terminated at any time. Agents must handle this gracefully.
187
+
188
+ ### The Challenge
189
+
190
+ ```
191
+ User starts research agent
192
+
193
+ Agent begins web search
194
+
195
+ User switches to another app
196
+
197
+ iOS suspends your app
198
+
199
+ Agent is mid-execution... what happens?
200
+ ```
201
+
202
+ ### Checkpoint/Resume Pattern
203
+
204
+ Save agent state before backgrounding, restore on foreground:
205
+
206
+ ```swift
207
+ class AgentOrchestrator: ObservableObject {
208
+ @Published var activeSessions: [AgentSession] = []
209
+
210
+ // Called when app is about to background
211
+ func handleAppWillBackground() {
212
+ for session in activeSessions {
213
+ saveCheckpoint(session)
214
+ session.transition(to: .backgrounded)
215
+ }
216
+ }
217
+
218
+ // Called when app returns to foreground
219
+ func handleAppDidForeground() {
220
+ for session in activeSessions where session.state == .backgrounded {
221
+ if let checkpoint = loadCheckpoint(session.id) {
222
+ resumeFromCheckpoint(session, checkpoint)
223
+ }
224
+ }
225
+ }
226
+
227
+ private func saveCheckpoint(_ session: AgentSession) {
228
+ let checkpoint = AgentCheckpoint(
229
+ sessionId: session.id,
230
+ conversationHistory: session.messages,
231
+ pendingToolCalls: session.pendingToolCalls,
232
+ partialResults: session.partialResults,
233
+ timestamp: Date()
234
+ )
235
+ storage.save(checkpoint, for: session.id)
236
+ }
237
+
238
+ private func resumeFromCheckpoint(_ session: AgentSession, _ checkpoint: AgentCheckpoint) {
239
+ session.messages = checkpoint.conversationHistory
240
+ session.pendingToolCalls = checkpoint.pendingToolCalls
241
+
242
+ // Resume execution if there were pending tool calls
243
+ if !checkpoint.pendingToolCalls.isEmpty {
244
+ session.transition(to: .running)
245
+ Task { await executeNextTool(session) }
246
+ }
247
+ }
248
+ }
249
+ ```
250
+
251
+ ### State Machine for Agent Lifecycle
252
+
253
+ ```swift
254
+ enum AgentState {
255
+ case idle // Not running
256
+ case running // Actively executing
257
+ case waitingForUser // Paused, waiting for user input
258
+ case backgrounded // App backgrounded, state saved
259
+ case completed // Finished successfully
260
+ case failed(Error) // Finished with error
261
+ }
262
+
263
+ class AgentSession: ObservableObject {
264
+ @Published var state: AgentState = .idle
265
+
266
+ func transition(to newState: AgentState) {
267
+ let validTransitions: [AgentState: Set<AgentState>] = [
268
+ .idle: [.running],
269
+ .running: [.waitingForUser, .backgrounded, .completed, .failed],
270
+ .waitingForUser: [.running, .backgrounded],
271
+ .backgrounded: [.running, .completed],
272
+ ]
273
+
274
+ guard validTransitions[state]?.contains(newState) == true else {
275
+ logger.warning("Invalid transition: \(state) → \(newState)")
276
+ return
277
+ }
278
+
279
+ state = newState
280
+ }
281
+ }
282
+ ```
283
+
284
+ ### Background Task Extension (iOS)
285
+
286
+ Request extra time when backgrounded during critical operations:
287
+
288
+ ```swift
289
+ class AgentOrchestrator {
290
+ private var backgroundTask: UIBackgroundTaskIdentifier = .invalid
291
+
292
+ func handleAppWillBackground() {
293
+ // Request extra time for saving state
294
+ backgroundTask = UIApplication.shared.beginBackgroundTask { [weak self] in
295
+ self?.endBackgroundTask()
296
+ }
297
+
298
+ // Save all checkpoints
299
+ Task {
300
+ for session in activeSessions {
301
+ await saveCheckpoint(session)
302
+ }
303
+ endBackgroundTask()
304
+ }
305
+ }
306
+
307
+ private func endBackgroundTask() {
308
+ if backgroundTask != .invalid {
309
+ UIApplication.shared.endBackgroundTask(backgroundTask)
310
+ backgroundTask = .invalid
311
+ }
312
+ }
313
+ }
314
+ ```
315
+
316
+ ### User Communication
317
+
318
+ Let users know what's happening:
319
+
320
+ ```swift
321
+ struct AgentStatusView: View {
322
+ @ObservedObject var session: AgentSession
323
+
324
+ var body: some View {
325
+ switch session.state {
326
+ case .backgrounded:
327
+ Label("Paused (app in background)", systemImage: "pause.circle")
328
+ .foregroundColor(.orange)
329
+ case .running:
330
+ Label("Working...", systemImage: "ellipsis.circle")
331
+ .foregroundColor(.blue)
332
+ case .waitingForUser:
333
+ Label("Waiting for your input", systemImage: "person.circle")
334
+ .foregroundColor(.green)
335
+ // ...
336
+ }
337
+ }
338
+ }
339
+ ```
340
+ </background_execution>
341
+
342
+ <permissions>
343
+ ## Permission Handling
344
+
345
+ Mobile agents may need access to system resources. Handle permission requests gracefully.
346
+
347
+ ### Common Permissions
348
+
349
+ | Resource | iOS Permission | Use Case |
350
+ |----------|---------------|----------|
351
+ | Photo Library | PHPhotoLibrary | Profile generation from photos |
352
+ | Files | Document picker | Reading user documents |
353
+ | Camera | AVCaptureDevice | Scanning book covers |
354
+ | Location | CLLocationManager | Location-aware recommendations |
355
+ | Network | (automatic) | Web search, API calls |
356
+
357
+ ### Permission-Aware Tools
358
+
359
+ Check permissions before executing:
360
+
361
+ ```swift
362
+ struct PhotoTools {
363
+ static func readPhotos() -> AgentTool {
364
+ tool(
365
+ name: "read_photos",
366
+ description: "Read photos from the user's photo library",
367
+ parameters: [
368
+ "limit": .number("Maximum photos to read"),
369
+ "dateRange": .string("Date range filter").optional()
370
+ ],
371
+ execute: { params, context in
372
+ // Check permission first
373
+ let status = await PHPhotoLibrary.requestAuthorization(for: .readWrite)
374
+
375
+ switch status {
376
+ case .authorized, .limited:
377
+ // Proceed with reading photos
378
+ let photos = await fetchPhotos(params)
379
+ return ToolResult(text: "Found \(photos.count) photos", images: photos)
380
+
381
+ case .denied, .restricted:
382
+ return ToolResult(
383
+ text: "Photo access needed. Please grant permission in Settings → Privacy → Photos.",
384
+ isError: true
385
+ )
386
+
387
+ case .notDetermined:
388
+ return ToolResult(
389
+ text: "Photo permission required. Please try again.",
390
+ isError: true
391
+ )
392
+
393
+ @unknown default:
394
+ return ToolResult(text: "Unknown permission status", isError: true)
395
+ }
396
+ }
397
+ )
398
+ }
399
+ }
400
+ ```
401
+
402
+ ### Graceful Degradation
403
+
404
+ When permissions aren't granted, offer alternatives:
405
+
406
+ ```swift
407
+ func readPhotos() async -> ToolResult {
408
+ let status = PHPhotoLibrary.authorizationStatus(for: .readWrite)
409
+
410
+ switch status {
411
+ case .denied, .restricted:
412
+ // Suggest alternative
413
+ return ToolResult(
414
+ text: """
415
+ I don't have access to your photos. You can either:
416
+ 1. Grant access in Settings → Privacy → Photos
417
+ 2. Share specific photos directly in our chat
418
+
419
+ Would you like me to help with something else instead?
420
+ """,
421
+ isError: false // Not a hard error, just a limitation
422
+ )
423
+ // ...
424
+ }
425
+ }
426
+ ```
427
+
428
+ ### Permission Request Timing
429
+
430
+ Don't request permissions until needed:
431
+
432
+ ```swift
433
+ // BAD: Request all permissions at launch
434
+ func applicationDidFinishLaunching() {
435
+ requestPhotoAccess()
436
+ requestCameraAccess()
437
+ requestLocationAccess()
438
+ // User is overwhelmed with permission dialogs
439
+ }
440
+
441
+ // GOOD: Request when the feature is used
442
+ tool("analyze_book_cover", async ({ image }) => {
443
+ // Only request camera access when user tries to scan a cover
444
+ let status = await AVCaptureDevice.requestAccess(for: .video)
445
+ if status {
446
+ return await scanCover(image)
447
+ } else {
448
+ return ToolResult(text: "Camera access needed for book scanning")
449
+ }
450
+ })
451
+ ```
452
+ </permissions>
453
+
454
+ <cost_awareness>
455
+ ## Cost-Aware Design
456
+
457
+ Mobile users may be on cellular data or concerned about API costs. Design agents to be efficient.
458
+
459
+ ### Model Tier Selection
460
+
461
+ Use the cheapest model that achieves the outcome:
462
+
463
+ ```swift
464
+ enum ModelTier {
465
+ case fast // gemini-flash: ~$0.25/1M tokens
466
+ case balanced // gemini-pro: ~$3/1M tokens
467
+ case powerful // gemini-ultra: ~$15/1M tokens
468
+
469
+ var modelId: String {
470
+ switch self {
471
+ case .fast: return "gemini-flash-20240307"
472
+ case .balanced: return "gemini-pro-20240229"
473
+ case .powerful: return "gemini-ultra-20240229"
474
+ }
475
+ }
476
+ }
477
+
478
+ // Match model to task complexity
479
+ let agentConfigs: [AgentType: ModelTier] = [
480
+ .quickLookup: .fast, // "What's in my library?"
481
+ .chatAssistant: .balanced, // General conversation
482
+ .researchAgent: .balanced, // Web search + synthesis
483
+ .profileGenerator: .powerful, // Complex photo analysis
484
+ .introductionWriter: .balanced,
485
+ ]
486
+ ```
487
+
488
+ ### Token Budgets
489
+
490
+ Limit tokens per agent session:
491
+
492
+ ```swift
493
+ struct AgentConfig {
494
+ let modelTier: ModelTier
495
+ let maxInputTokens: Int
496
+ let maxOutputTokens: Int
497
+ let maxTurns: Int
498
+
499
+ static let research = AgentConfig(
500
+ modelTier: .balanced,
501
+ maxInputTokens: 50_000,
502
+ maxOutputTokens: 4_000,
503
+ maxTurns: 20
504
+ )
505
+
506
+ static let quickChat = AgentConfig(
507
+ modelTier: .fast,
508
+ maxInputTokens: 10_000,
509
+ maxOutputTokens: 1_000,
510
+ maxTurns: 5
511
+ )
512
+ }
513
+
514
+ class AgentSession {
515
+ var totalTokensUsed: Int = 0
516
+
517
+ func checkBudget() -> Bool {
518
+ if totalTokensUsed > config.maxInputTokens {
519
+ transition(to: .failed(AgentError.budgetExceeded))
520
+ return false
521
+ }
522
+ return true
523
+ }
524
+ }
525
+ ```
526
+
527
+ ### Network-Aware Execution
528
+
529
+ Defer heavy operations to WiFi:
530
+
531
+ ```swift
532
+ class NetworkMonitor: ObservableObject {
533
+ @Published var isOnWiFi: Bool = false
534
+ @Published var isExpensive: Bool = false // Cellular or hotspot
535
+
536
+ private let monitor = NWPathMonitor()
537
+
538
+ func startMonitoring() {
539
+ monitor.pathUpdateHandler = { [weak self] path in
540
+ DispatchQueue.main.async {
541
+ self?.isOnWiFi = path.usesInterfaceType(.wifi)
542
+ self?.isExpensive = path.isExpensive
543
+ }
544
+ }
545
+ monitor.start(queue: .global())
546
+ }
547
+ }
548
+
549
+ class AgentOrchestrator {
550
+ @ObservedObject var network = NetworkMonitor()
551
+
552
+ func startResearchAgent(for book: Book) async {
553
+ if network.isExpensive {
554
+ // Warn user or defer
555
+ let proceed = await showAlert(
556
+ "Research uses data",
557
+ message: "This will use approximately 1-2 MB of cellular data. Continue?"
558
+ )
559
+ if !proceed { return }
560
+ }
561
+
562
+ // Proceed with research
563
+ await runAgent(ResearchAgent.create(book: book))
564
+ }
565
+ }
566
+ ```
567
+
568
+ ### Batch API Calls
569
+
570
+ Combine multiple small requests:
571
+
572
+ ```swift
573
+ // BAD: Many small API calls
574
+ for book in books {
575
+ await agent.chat("Summarize \(book.title)")
576
+ }
577
+
578
+ // GOOD: Batch into one request
579
+ let bookList = books.map { $0.title }.joined(separator: ", ")
580
+ await agent.chat("Summarize each of these books briefly: \(bookList)")
581
+ ```
582
+
583
+ ### Caching
584
+
585
+ Cache expensive operations:
586
+
587
+ ```swift
588
+ class ResearchCache {
589
+ private var cache: [String: CachedResearch] = [:]
590
+
591
+ func getCachedResearch(for bookId: String) -> CachedResearch? {
592
+ guard let cached = cache[bookId] else { return nil }
593
+
594
+ // Expire after 24 hours
595
+ if Date().timeIntervalSince(cached.timestamp) > 86400 {
596
+ cache.removeValue(forKey: bookId)
597
+ return nil
598
+ }
599
+
600
+ return cached
601
+ }
602
+
603
+ func cacheResearch(_ research: Research, for bookId: String) {
604
+ cache[bookId] = CachedResearch(
605
+ research: research,
606
+ timestamp: Date()
607
+ )
608
+ }
609
+ }
610
+
611
+ // In research tool
612
+ tool("web_search", async ({ query, bookId }) => {
613
+ // Check cache first
614
+ if let cached = cache.getCachedResearch(for: bookId) {
615
+ return ToolResult(text: cached.research.summary, cached: true)
616
+ }
617
+
618
+ // Otherwise, perform search
619
+ let results = await webSearch(query)
620
+ cache.cacheResearch(results, for: bookId)
621
+ return ToolResult(text: results.summary)
622
+ })
623
+ ```
624
+
625
+ ### Cost Visibility
626
+
627
+ Show users what they're spending:
628
+
629
+ ```swift
630
+ struct AgentCostView: View {
631
+ @ObservedObject var session: AgentSession
632
+
633
+ var body: some View {
634
+ VStack(alignment: .leading) {
635
+ Text("Session Stats")
636
+ .font(.headline)
637
+
638
+ HStack {
639
+ Label("\(session.turnCount) turns", systemImage: "arrow.2.squarepath")
640
+ Spacer()
641
+ Label(formatTokens(session.totalTokensUsed), systemImage: "text.word.spacing")
642
+ }
643
+
644
+ if let estimatedCost = session.estimatedCost {
645
+ Text("Est. cost: \(estimatedCost, format: .currency(code: "USD"))")
646
+ .font(.caption)
647
+ .foregroundColor(.secondary)
648
+ }
649
+ }
650
+ }
651
+ }
652
+ ```
653
+ </cost_awareness>
654
+
655
+ <offline_handling>
656
+ ## Offline Graceful Degradation
657
+
658
+ Handle offline scenarios gracefully:
659
+
660
+ ```swift
661
+ class ConnectivityAwareAgent {
662
+ @ObservedObject var network = NetworkMonitor()
663
+
664
+ func executeToolCall(_ toolCall: ToolCall) async -> ToolResult {
665
+ // Check if tool requires network
666
+ let requiresNetwork = ["web_search", "web_fetch", "call_api"]
667
+ .contains(toolCall.name)
668
+
669
+ if requiresNetwork && !network.isConnected {
670
+ return ToolResult(
671
+ text: """
672
+ I can't access the internet right now. Here's what I can do offline:
673
+ - Read your library and existing research
674
+ - Answer questions from cached data
675
+ - Write notes and drafts for later
676
+
677
+ Would you like me to try something that works offline?
678
+ """,
679
+ isError: false
680
+ )
681
+ }
682
+
683
+ return await executeOnline(toolCall)
684
+ }
685
+ }
686
+ ```
687
+
688
+ ### Offline-First Tools
689
+
690
+ Some tools should work entirely offline:
691
+
692
+ ```swift
693
+ let offlineTools: Set<String> = [
694
+ "read_file",
695
+ "write_file",
696
+ "list_files",
697
+ "read_library", // Local database
698
+ "search_local", // Local search
699
+ ]
700
+
701
+ let onlineTools: Set<String> = [
702
+ "web_search",
703
+ "web_fetch",
704
+ "publish_to_cloud",
705
+ ]
706
+
707
+ let hybridTools: Set<String> = [
708
+ "publish_to_feed", // Works offline, syncs later
709
+ ]
710
+ ```
711
+
712
+ ### Queued Actions
713
+
714
+ Queue actions that require connectivity:
715
+
716
+ ```swift
717
+ class OfflineQueue: ObservableObject {
718
+ @Published var pendingActions: [QueuedAction] = []
719
+
720
+ func queue(_ action: QueuedAction) {
721
+ pendingActions.append(action)
722
+ persist()
723
+ }
724
+
725
+ func processWhenOnline() {
726
+ network.$isConnected
727
+ .filter { $0 }
728
+ .sink { [weak self] _ in
729
+ self?.processPendingActions()
730
+ }
731
+ }
732
+
733
+ private func processPendingActions() {
734
+ for action in pendingActions {
735
+ Task {
736
+ try await execute(action)
737
+ remove(action)
738
+ }
739
+ }
740
+ }
741
+ }
742
+ ```
743
+ </offline_handling>
744
+
745
+ <battery_awareness>
746
+ ## Battery-Aware Execution
747
+
748
+ Respect device battery state:
749
+
750
+ ```swift
751
+ class BatteryMonitor: ObservableObject {
752
+ @Published var batteryLevel: Float = 1.0
753
+ @Published var isCharging: Bool = false
754
+ @Published var isLowPowerMode: Bool = false
755
+
756
+ var shouldDeferHeavyWork: Bool {
757
+ return batteryLevel < 0.2 && !isCharging
758
+ }
759
+
760
+ func startMonitoring() {
761
+ UIDevice.current.isBatteryMonitoringEnabled = true
762
+
763
+ NotificationCenter.default.addObserver(
764
+ forName: UIDevice.batteryLevelDidChangeNotification,
765
+ object: nil,
766
+ queue: .main
767
+ ) { [weak self] _ in
768
+ self?.batteryLevel = UIDevice.current.batteryLevel
769
+ }
770
+
771
+ NotificationCenter.default.addObserver(
772
+ forName: NSNotification.Name.NSProcessInfoPowerStateDidChange,
773
+ object: nil,
774
+ queue: .main
775
+ ) { [weak self] _ in
776
+ self?.isLowPowerMode = ProcessInfo.processInfo.isLowPowerModeEnabled
777
+ }
778
+ }
779
+ }
780
+
781
+ class AgentOrchestrator {
782
+ @ObservedObject var battery = BatteryMonitor()
783
+
784
+ func startAgent(_ config: AgentConfig) async {
785
+ if battery.shouldDeferHeavyWork && config.isHeavy {
786
+ let proceed = await showAlert(
787
+ "Low Battery",
788
+ message: "This task uses significant battery. Continue or defer until charging?"
789
+ )
790
+ if !proceed { return }
791
+ }
792
+
793
+ // Adjust model tier based on battery
794
+ let adjustedConfig = battery.isLowPowerMode
795
+ ? config.withModelTier(.fast)
796
+ : config
797
+
798
+ await runAgent(adjustedConfig)
799
+ }
800
+ }
801
+ ```
802
+ </battery_awareness>
803
+
804
+ <on_device_vs_cloud>
805
+ ## On-Device vs. Cloud
806
+
807
+ Understanding what runs where in a mobile agent-native app:
808
+
809
+ | Component | On-Device | Cloud |
810
+ |-----------|-----------|-------|
811
+ | Orchestration | ✅ | |
812
+ | Tool execution | ✅ (file ops, photo access, HealthKit) | |
813
+ | LLM calls | | ✅ (Antigravity API) |
814
+ | Checkpoints | ✅ (local files) | Optional via iCloud |
815
+ | Long-running agents | Limited by iOS | Possible with server |
816
+
817
+ ### Implications
818
+
819
+ **Network required for reasoning:**
820
+ - The app needs network connectivity for LLM calls
821
+ - Design tools to degrade gracefully when network is unavailable
822
+ - Consider offline caching for common queries
823
+
824
+ **Data stays local:**
825
+ - File operations happen on device
826
+ - Sensitive data never leaves the device unless explicitly synced
827
+ - Privacy is preserved by default
828
+
829
+ **Long-running agents:**
830
+ For truly long-running agents (hours), consider a server-side orchestrator that can run indefinitely, with the mobile app as a viewer and input mechanism.
831
+ </on_device_vs_cloud>
832
+
833
+ <checklist>
834
+ ## Mobile Agent-Native Checklist
835
+
836
+ **iOS Storage:**
837
+ - [ ] iCloud Documents as primary storage (or conscious alternative)
838
+ - [ ] Local Documents fallback when iCloud unavailable
839
+ - [ ] Handle `.icloud` placeholder files (trigger download)
840
+ - [ ] Use NSFileCoordinator for conflict-safe writes
841
+
842
+ **Background Execution:**
843
+ - [ ] Checkpoint/resume implemented for all agent sessions
844
+ - [ ] State machine for agent lifecycle (idle, running, backgrounded, etc.)
845
+ - [ ] Background task extension for critical saves (30 second window)
846
+ - [ ] User-visible status for backgrounded agents
847
+
848
+ **Permissions:**
849
+ - [ ] Permissions requested only when needed, not at launch
850
+ - [ ] Graceful degradation when permissions denied
851
+ - [ ] Clear error messages with Settings deep links
852
+ - [ ] Alternative paths when permissions unavailable
853
+
854
+ **Cost Awareness:**
855
+ - [ ] Model tier matched to task complexity
856
+ - [ ] Token budgets per session
857
+ - [ ] Network-aware (defer heavy work to WiFi)
858
+ - [ ] Caching for expensive operations
859
+ - [ ] Cost visibility to users
860
+
861
+ **Offline Handling:**
862
+ - [ ] Offline-capable tools identified
863
+ - [ ] Graceful degradation for online-only features
864
+ - [ ] Action queue for sync when online
865
+ - [ ] Clear user communication about offline state
866
+
867
+ **Battery Awareness:**
868
+ - [ ] Battery monitoring for heavy operations
869
+ - [ ] Low power mode detection
870
+ - [ ] Defer or downgrade based on battery state
871
+ </checklist>