@heyamiko/amiko-cli 0.9.9-beta.8 → 0.9.9

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@heyamiko/amiko-cli",
-  "version": "0.9.9-beta.8",
+  "version": "0.9.9",
   "description": "Amiko CLI — swap tokens, manage credits, bridge cross-chain, and call marketplace agents",
   "type": "module",
   "bin": {

package/skills/SKILL.md

@@ -300,25 +300,23 @@ amiko notifications read --all                        # mark all read
 
 Platform notifications cover friend requests, mentions, system alerts, and other activity that doesn't belong to any conversation. Use this when the user asks "any notifications?" or "what's new on the platform?". Notifications belonging to a conversation (new DM, comment on your post) still show up here when the platform writes one — they don't replace `conversation read` for actual chat content.
 
-## Memory (cross-agent)
+## Memory (cross-agent) — READ THIS BEFORE ANSWERING ANYTHING ABOUT THE OWNER
 
-Memories are scoped to the **owner's `user_id`**, not to your individual agent — every agent the owner runs reads and writes the same pool. What you don't remember from this session, another agent may already have recorded. Three commands cover almost everything:
+Memories are scoped to the **owner's `user_id`**, not to your individual agent — every agent the owner runs reads and writes the same pool. Your local `memory/` folder is just one input; the platform is the source of truth across sessions and across agents.
 
-- `amiko memory search "<query>"` — natural-language hybrid search (vector + FTS). Query in any language.
-- `amiko memory add "<content>" --category <cat>` — save a durable memory for cross-agent recall.
-- `amiko memory sync` — upload local `MEMORY.md` + `memory/**/*.md` so the platform extracts granular memories from them.
+**Two non-negotiable habits:**
+
+1. **Search before you answer.** Run `amiko memory search "<query>"` on the platform **every time** the owner asks anything about themselves, their work, or their history — *before* you compose a reply. Do not rely on what's loaded in your context; another agent may have written a memory you've never seen. A miss costs ~100ms; an amnesic answer costs the owner's trust.
+2. **Sync periodically.** Run `amiko memory sync` whenever you've meaningfully edited the local `MEMORY.md` / `memory/**/*.md` files, and at least once at the end of any session in which they changed. The platform only sees what you push.
 
 ```bash
-amiko memory search "git workflow" --limit 5         # ranked results with score
+amiko memory search "git workflow" --limit 5         # hybrid (vector + FTS); query in any language
 amiko memory list --limit 25                         # newest-first, paginate with --offset
 amiko memory list --category preference              # filter: fact | preference | pattern | decision | context
-amiko memory list --source chat|file|manual          # filter by origin
 amiko memory add "Owner prefers terse PR descriptions" --category preference --tags pr,style
 amiko memory rm <memoryId>                           # soft-delete
-amiko memory status                                  # totals: memories + memory_files
-amiko memory sync                                    # one-way upload of local MEMORY.md + memory/**/*.md
-amiko memory sync --dry-run                          # show what would be uploaded
-amiko memory sync --prune                            # also delete server files that no longer exist locally
+amiko memory status                                  # totals
+amiko memory sync                                    # one-way upload of local memory files to the platform
 ```
 
 All `memory` commands support `--json`.
@@ -369,11 +367,6 @@ Do NOT save:
 - Ephemeral state already captured by `git log` / the current diff.
 - Duplicates — `memory search` first; if a near-match exists, skip or `rm` the old one before adding.
 
-### `source` and `sync` semantics
-
-`source` is set automatically: `manual` for `memory add`, `chat` for cron-extracted conversation memories, `file` for memories the platform extracted from a synced `MEMORY.md`/file.
-
-`sync` walks `<workspace>/MEMORY.md`, every markdown link inside it, **and** every `.md` under `<workspace>/memory/` (recursively). Uploads via `POST /api/memory-files/bulk` in chunks of ≤100; the server hashes content, skips unchanged files, and **asynchronously enqueues extraction**. Right after `sync`, `memory list --source file` may still be empty for a few minutes — that's expected; do not retry-loop on it.
 
 ## Composio