neo-agent-skills 0.1.63 → 0.1.65

package/.agents/skills/memory-mining/references/memory-mining-protocol.md

@@ -35,10 +35,10 @@ Do not issue one query with many keywords. Issue 2–4 queries with **different
 
 Choose the first memory surface by freshness, not by habit.
 
-1. **Recent / active-session investigations:** run **`query_raw_memories` first or in parallel with `query_summaries`** when the event is same-day, just happened, active-ticket / active-PR bound, wake/A2A coordination, or otherwise likely to live in unsummarized turns. `add_memory()` can persist raw turns before any session summary exists, so a summary miss is not a memory miss until a freshness-shaped raw query also misses.
-2. **Older / stable history:** run **`query_summaries` first** for broad conceptual exploration, older decisions, and low-freshness historical context. Session-level topics surface cheaply when summarization has had time to digest the work.
-3. **`query_raw_memories` after a summary hit:** use raw memories to recover turn-level reasoning, exact commands, message IDs, or decision texture behind a relevant summary.
-4. **`ask_knowledge_base`** — only when the answer lives in indexed code/guides, not in session memory. If unsure, run memory first; KB is cheaper to skip than memory is. This is the Memory Core analogue of the GitHub/KB freshness gap corrected in #10646.
+1. **Recent / active-session investigations:** run **`query_raw_memories` (`detail: 'summary'`, plus `agentIdentity` when author-scoped) first or in parallel with `query_summaries`** when the event is same-day, active-ticket / active-PR bound, wake/A2A coordination, or likely in unsummarized turns (`add_memory()` persists raw turns before any session summary exists).
+2. **Older / stable history:** run **`query_summaries` first** (with `agentIdentity` when author-scoped) for broad conceptual exploration and older decisions.
+3. **`query_raw_memories` (`detail: 'full'`) after a summary hit:** hydrate matching turns (`projection: 'public'` on cross-seat team reads) to recover turn-level reasoning, commands, message IDs, or decision texture.
+4. **`ask_knowledge_base`** — only when the answer lives in indexed code/guides, not in session memory. If unsure, run memory first (#10646).
 
 ### Good vs bad query shapes
 

package/.agents/skills/neo-app-authoring/SKILL.md

@@ -3,4 +3,4 @@ name: neo-app-authoring
 description: "Write app code in a neo.mjs workspace the engine's way: classes, reactive configs, stores, state providers, SCSS. Triggers: before writing or changing anything under apps/ in a workspace built on neo.mjs."
 ---
 # Neo App Authoring
-Before writing or changing application code in a neo.mjs workspace, you MUST use the `view_file` tool to read and follow `.agents/skills/neo-app-authoring/references/app-authoring-guide.md`.
+Before writing or changing application code in a neo.mjs workspace, you MUST read and follow `.agents/skills/neo-app-authoring/references/app-authoring-guide.md`.

package/.agents/skills/neural-link/SKILL.md

@@ -3,4 +3,4 @@ name: neural-link
 description: "Tactical guide for Neural Link MCP tools on live Neo.mjs apps; use when asked to interact with the browser, inspect UI natively, patch live code, or use the Neural Link."
 ---
 # Neural Link Workflow
-If you are tasked with debugging, monitoring, or modifying the live Neo.mjs runtime environment, you MUST immediately use the `view_file` tool to read and strictly adhere to `.agents/skills/neural-link/references/operational-handbook.md` before invoking any native MCP tools.
+If you are tasked with debugging, monitoring, or modifying the live Neo.mjs runtime environment, you MUST immediately read and strictly adhere to `.agents/skills/neural-link/references/operational-handbook.md` before invoking any native MCP tools.

package/.agents/skills/neural-link/references/operational-handbook.md

@@ -4,6 +4,13 @@ Your task is to effectively query and mutate a running Neo.mjs application using
 
 **DO NOT GUESS.** Without specific chains of verification, you will inject corrupted instructions or patch incorrect classes. You MUST follow these sequences.
 
+## 0. Getting Connected
+
+- The MCP server talks to a local Bridge on `127.0.0.1:8081` (`NEO_NL_PORT` overrides it) and starts one when none runs. The app's page dials that Bridge in development (`useAiClient` in its `neo-config.json`).
+- Zero `sessions` in `healthcheck`, or "No active App Worker sessions found", means no page is connected: open the app's page. One that loaded before the Bridge started stops dialing after five tries (about 15 seconds) and dials again only when its window is focused or shown, never on a timer: focus or reload it.
+- `get_mcp_tool_handbook({toolId})` returns a tool's usage guide and the parameter documentation its compact description and schema leave out; read it before an unfamiliar call.
+- If `get_window_topology` lists no window although a session exists (engines up to 13.3.2 can miss a window whose page loaded after the Bridge started), take the `windowId` from `query_component`'s result.
+
 ## 1. The Discovery Chain (UI Inspection)
 
 When asked to locate or verify an element on the screen:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "neo-agent-skills",
-  "version": "0.1.63",
+  "version": "0.1.65",
   "description": "Install the workflow substrate a cross-model AI team actually runs on: ticket intake, PR review gates, lane coordination and guards, versioned as a package. The skills Neo.mjs maintainers load every session — materialized into your repo, not copied.",
   "type": "module",
   "license": "MIT",