@codeursenior/pek 0.1.10 → 0.1.11

package/README.md

@@ -20,10 +20,13 @@ Replace `npx @codeursenior/boyscout ui [folder]` with `npx @codeursenior/pek aud
 
 ## Explore
 
+Token Optimizer is the home page. Return to it by clicking the avatar, agent name, or factory path. Rules and Optimization checklist are temporarily hidden from navigation.
+
+- **Agent filter:** select one client using the icons below the avatar. The same selection applies to Context, Rules, Skills (including Whiteboard), MCP servers, and Token Optimizer. Click the selected icon again to show all agents. All agents are selected at startup. The project/user scope filter is restored from localStorage in this browser.
 - **Context:** interactive graph with pan, zoom, draggable nodes, and file previews. Solid lines are explicit references; dashed lines show nested instruction scopes. Switch to Files for a table with line counts and client discovery. Filter by coding agent in either view.
 - **Rules:** searchable list of dedicated Cursor, Claude Code, and GitHub Copilot instruction files, with source paths, declared file conditions, client discovery, and previews. Metadata does not prove a rule loaded in a session.
 - **Skills:** names, descriptions, invocation policy, source paths, and discovery indicators for Cursor, Claude Code, Codex, and GitHub Copilot in VS Code. Identical copies in the same scope are grouped, retaining their paths.
-- **Skills Whiteboard:** switch from Table to Whiteboard to arrange every skill as a draggable card, connect your workflow with arrows, and add movable text titles. Solid blue arrows show explicit references in SKILL.md and cannot be edited. Dashed amber arrows are your local drawings: select one to delete it, drag its endpoints onto other cards, or change its From and To menus. All skills stay visible, including unused ones. Select a card and choose Open skill to inspect it; arrow keys move focused cards and titles. Pan, zoom and Fit skills help navigate larger inventories.
+- **Skills Whiteboard:** switch from Files to Whiteboard to arrange every skill as a draggable card, connect your workflow with arrows, and add movable text titles. Solid blue arrows show explicit references in SKILL.md and cannot be edited. Dashed amber arrows are your local drawings: select one to delete it, drag its endpoints onto other cards, or change its From and To menus. All skills discoverable by the selected agent stay visible, including unused ones. Select a card and choose Open skill to inspect it; arrow keys move focused cards and titles. Pan, zoom and Fit skills help navigate larger inventories. Wheel and trackpad zoom stays anchored under the pointer and scales gradually with scroll distance; the zoom buttons use the center of the canvas. Use Fullscreen to expand the board across the browser viewport; Exit fullscreen or Escape restores the page without changing the composition.
 - **MCP servers:** names, configuration sources, transport, and client indicators. The tool reads configuration without launching servers or transmitting credentials.
 - **Project and User:** green and purple distinguish assets in the selected folder from assets in supported home-directory locations. Scope filters work in all views.
 
@@ -43,7 +46,7 @@ npx @codeursenior/pek audit agent --port 4317 --no-open
 
 The default loopback port is **4242**, a nod to the answer in The Hitchhiker's Guide to the Galaxy. If it is occupied, close the other instance or pass `--port <number>`; the CLI does not silently change ports. Use `--port 0` to request an available port. The terminal URL includes a session token. Keep that URL private while the viewer is running.
 
-Whiteboard positions, manual arrows and titles are automatically saved in **localStorage for each project**. They are local to this browser and port, and are lost if you clear its site data. Use the same browser and port to restore your board at the next launch. Switching between Table and Whiteboard keeps the composition. List filters do not hide cards on the whiteboard. Inventory and source arrows update only at startup or Rescan; added skills appear beside the existing layout, while removed skills and their attached arrows leave the board. Editing a skill's content preserves its position. No skill files are changed or executed by whiteboard drawings. File-based sharing such as `skill-map.json` is not implemented.
+Whiteboard positions, manual arrows and titles are automatically saved in **localStorage for each project**. They are local to this browser and port, and are lost if you clear its site data. Use the same browser and port to restore your board at the next launch. Switching between Files and Whiteboard keeps the composition. The global agent filter hides unrelated cards and their arrows without changing saved positions, titles, or drawings. Scope and search filters in the table do not hide whiteboard cards. Inventory and source arrows update only at startup or Rescan; added skills appear beside the existing layout, while removed skills and their attached arrows leave the board. Editing a skill's content preserves its position. No skill files are changed or executed by whiteboard drawings. File-based sharing such as `skill-map.json` is not implemented.
 
 ## Optional agent identity
 
@@ -59,7 +62,7 @@ Give the agent a name and avatar in the sidebar by adding either or both files t
 
 Use `AVATAR.png` or `AVATAR.jpeg` for the image, up to 2 MiB. The file signature must match its extension. PNG takes priority when both files pass those checks. The image is cropped to a circle; if the browser cannot decode it, the default icon remains.
 
-These are optional, independent settings: without a valid name, the sidebar still says **boyscout**; without an avatar, it keeps its default icon. The workspace name and path stay unchanged. Click **Rescan** after adding, changing or removing either file.
+These are optional, independent settings: without a valid name, the sidebar uses the scanned folder name with its first letter capitalized; without an avatar, it keeps its default icon. The factory path appears below Agent Visualizer. Click **Rescan** after adding, changing or removing either file.
 
 This is an internal Agent Visualizer convention, not an ecosystem standard or a configuration for other coding agents. Only `.agents/` directly under the selected root is used, with the exact filenames above; nested projects and home-directory identities are not inherited. Symlinks outside the selected project are ignored. Images are delivered with the authenticated scan, without an external image request or a public file route.
 
@@ -81,7 +84,7 @@ Codex behavior instructions in `AGENTS.md` stay in Context. Codex `.codex/rules/
 
 Client icons show **file-based discovery**, not whether an application is installed, whether a skill's dependencies are available, or whether an asset is active in a particular conversation. Nested skills may only load when that client works in their folder. A skill outside recognized directories is listed with inactive client icons.
 
-Claude/Cursor/Copilot `disable-model-invocation` and Claude/Copilot `user-invocable` frontmatter are read. Codex `agents/openai.yaml` policy `allow_implicit_invocation` is read separately. Hover an icon or open a skill to inspect per-client invocation. Legacy `.codex/skills` paths are identified in details.
+Claude/Cursor/Copilot `disable-model-invocation` and Claude/Copilot `user-invocable` frontmatter are read. Codex `agents/openai.yaml` policy `allow_implicit_invocation` is read separately. Hover an icon or open a skill to inspect per-client invocation. Legacy `.codex/skills` paths are identified in details. Bundled skills under `.codex/skills/.system/` are attributed only to Codex; ordinary skills in `.codex/skills/` remain discoverable by Cursor through its compatibility imports.
 
 ## Honest limits
 
@@ -129,4 +132,4 @@ MIT © Simon Dieny. Independent project, not affiliated with Cursor, Anthropic,
 
 ## Token Optimizer
 
-Open **Token Optimizer** to estimate one agent’s startup and on-demand context, compare one-load input costs using maintained model presets, and inspect the largest files and skills. The RTK checklist detects installation and provides manual installation instructions. These are static local estimates, not session usage or spending. See [the estimation method and limits](docs/token-optimizer.md).
+All agents are selected at startup. The home page estimates their combined startup and on-demand context, counting shared entries once. Select an agent below the avatar to narrow the estimate, compare one-load input costs using maintained model presets, and inspect the largest files and skills. The RTK checklist detects installation and provides manual installation instructions. These are static local estimates, not session usage or spending. See [the estimation method and limits](docs/token-optimizer.md).

package/dist/browser/index.html

@@ -5,8 +5,8 @@
     <meta name="viewport" content="width=device-width, initial-scale=1"/>
     <title>Agent Visualizer</title>
     <link rel="icon" href="/favicon.svg"/>
-  <link rel="stylesheet" href="styles-2QCIL5GS.css"></head>
+  <link rel="stylesheet" href="styles-AWH5SAZF.css"></head>
   <body>
     <app-root ngcspnonce="__CSP_NONCE__"></app-root>
-  <script src="main-ITCUKQ5X.js" type="module" nonce="__CSP_NONCE__"></script></body>
+  <script src="main-ZG5BUSN5.js" type="module" nonce="__CSP_NONCE__"></script></body>
 </html>