Menü
Detect, create, and manage the AI-Memory fleet communication channel. Fires on bootstrap, session start (announcements), and feedback writes.
| type | skill |
| lifecycle | stable |
| inheritance | inheritable |
| name | ai-memory-setup |
| description | Detect, create, and manage the AI-Memory fleet communication channel. Fires on bootstrap, session start (announcements), and feedback writes. |
| tier | standard |
| applyTo | **/AI-Memory/**,**/cognitive-config*,**/*feedback*,**/*announcement*,**/*fleet*,**/*memory-setup* |
| currency | "2026-05-02T00:00:00.000Z" |
| lastReviewed | "2026-05-02T00:00:00.000Z" |
AI-Memory is the shared folder on the user's cloud drive where ACT heirs exchange feedback, announcements, and registry data. This skill covers detection, creation, path resolution, and ongoing read/write operations.
| Provider | Folder patterns detected | Notes |
|---|---|---|
| OneDrive | OneDrive, OneDrive - * (personal, family, business) | Most common; multiple accounts create multiple folders |
| iCloud | iCloudDrive, iCloud Drive, iCloud~com~apple~CloudDocs | macOS and Windows variants |
| Dropbox | Dropbox, Dropbox (Personal), Dropbox (Business) | Personal and business variants |
| Google Drive | Google Drive, My Drive | Drive for Desktop sync folder |
| Box | Box, Box Sync | Enterprise cloud storage |
| MEGA | MEGA, MEGAsync | Encrypted cloud storage |
| pCloud | pCloud, pCloud Drive | European cloud storage |
| Nextcloud | Nextcloud | Self-hosted cloud |
Discovery scans the user's HOME directory for any folder matching these patterns. Unknown cloud drives are ignored (the user can still point to them manually via ai_memory_root).
Resolve the AI-Memory root in this order:
.github/config/cognitive-config.json. If ai_memory_root is set, use <HOME>/<ai_memory_root>/AI-Memory.<HOME> for folders matching known cloud provider patterns. Skip any folder listed in ai_memory_exclude. Among matches, prefer drives that already contain an AI-Memory/ subfolder.~/AI-Memory (no cloud sync, still functional)ai_memory_exclude in cognitive-config.json lists folder names to skip (e.g., ["OneDrive - Microsoft"] to avoid the work account).The resolution order matches _registry.cjs (the muscle used by bootstrap-heir.cjs and upgrade-self.cjs). LLM heirs and scripts must agree on the same path.
<cloud-drive>/AI-Memory/
README.md # Channel overview
feedback/
README.md
alex-act/ # Heir feedback inbox
README.md
announcements/
alex-act/ # Fleet-wide announcements
README.md
heirs/
registry.json # Fleet registry (auto-maintained)
knowledge/ # Shared knowledge base (see index.json)
insights/ # Analytical insights
AI-Memory/knowledge/ contains reference material installed from the Plugin Mall. These are not loaded into the brain -- they are consulted on demand when a task matches their domain.
Read AI-Memory/knowledge/index.json to discover available packages. Each entry has name, keywords, use_phase, and path. When a task involves a matching phase (planning, audit, review, implementation) or keyword, read the referenced reference.md for guidance.
This costs zero tokens until you actually read it.
On every session start, resolve the AI-Memory root. If found:
announcements/alex-act/ for unread filesIf not found: note it silently. Do not prompt the user unless they explicitly ask about fleet communication.
When AI-Memory is needed but doesn't exist (bootstrap, first feedback write):
Discover which cloud drives exist:
node .github/scripts/_registry.cjs --discover
Output lists all detected cloud drives with provider name and whether AI-Memory exists.
If multiple drives found, ask the user which one to use. If only one, use it. If none, offer ~/AI-Memory as local fallback.
Create the folder structure:
node .github/scripts/_registry.cjs --init "<drive-name>"
Or during bootstrap, use the --ai-memory flag:
node bootstrap-heir.cjs --target . --heir-id my-project --ai-memory "Dropbox" --apply
The choice is automatically persisted in .github/config/cognitive-config.json:
{
"ai_memory_root": "Dropbox",
"ai_memory_exclude": ["OneDrive - Microsoft"]
}
Verify: node .github/scripts/_registry.cjs --resolve . should return the AI-Memory path.
When the heir observes friction worth surfacing:
feedback/alex-act/YYYY-MM-DD-<short-slug>.mdcross-project-isolation.instructions.mdOn session start (triggered by greeting-checkin.instructions.md):
announcements/alex-act/ (skip README.md)Users may have multiple OneDrive accounts (personal + work). The ai_memory_exclude field prevents heirs from writing to the wrong drive.
| Scenario | Resolution |
|---|---|
| One OneDrive | Auto-detected, no config needed |
| Two OneDrives, AI-Memory in one | ai_memory_root pins the correct one |
| Two OneDrives, AI-Memory in both | ai_memory_exclude blocks the wrong one |
| No OneDrive, has iCloud | iCloud candidate resolves |
| No cloud drive at all | ~/AI-Memory local fallback |
cognitive-config.json fields (heir-owned, not overwritten on upgrade):
| Field | Type | Default | Purpose |
|---|---|---|---|
ai_memory_root | string | (auto-detect) | Cloud drive folder name to use |
ai_memory_exclude | string[] | [] | Folder names to skip during detection |
| Anti-pattern | Correction |
|---|---|
| Hardcoding an absolute path | Use the resolution algorithm; paths differ per user and OS |
Creating AI-Memory in OneDrive - Microsoft (work account) | Personal cloud drive only; work drives may have retention policies |
| Writing feedback without stripping project context | Always apply cross-project-isolation before writing |
| Reading feedback as a heir | Feedback is for the Supervisor; heirs read announcements only |
| Creating AI-Memory on every session | Create once on bootstrap; subsequent sessions just resolve |
The _registry.cjs script (shipped with Edition) provides both a programmatic API and a CLI:
| Function | Purpose |
|---|---|
resolveAiMemoryRoot(heirRoot?) | Returns the AI-Memory path or null; honors cognitive-config.json |
discoverCloudDrives(heirRoot?) | Scans HOME for cloud drive folders; returns [{ name, path, provider, hasAiMemory }] |
initAiMemory(driveName) | Creates full folder structure + READMEs |
upsertHeir(marker, repoPath) | Updates heirs/registry.json |
node .github/scripts/_registry.cjs --discover # List cloud drives
node .github/scripts/_registry.cjs --init "<name>" # Create AI-Memory in named drive
node .github/scripts/_registry.cjs --resolve [dir] # Resolve AI-Memory root
node .github/scripts/bootstrap-heir.cjs --ai-memory "<drive-name>" ...
Overrides auto-selection during bootstrap. Persists to cognitive-config.json.
End-to-end academic paper drafting for CHI, HBR, journals, and conferences with venue-specific templates, drafting workflows, and revision strategies.
Generate on-brand Alex — ACT Edition SVG banners for documents (READMEs, plans, notes, release artifacts)
Patterns for fiction, narrative structure, character development, dialogue, and storytelling craft.
Convert Word documents (.docx) to clean Markdown with image extraction and pandoc cleanup
Greeting-triggered self-check — recognise greetings, check Edition version against the upstream tag, scan AI-Memory announcements, and report inside the greeting reply
Write markdown that passes markdownlint on first attempt — encode the most common rules as muscle memory