| name | __platform__ |
| description | NAP workspace capabilities reference. Consult when working with browser automation, sandboxed code execution, file sharing between agents via /mnt/afs, persistent cross-session memory under /mnt/memory, creating or editing skills, proposing workspace configuration changes via Builder Mode, or (Codex) generated images. |
NAP Workspace Capabilities
The NAP platform provides several runtime facilities inside this workspace. Each capability has a dedicated reference file under reference/; read the matching one before using the capability, since exact tool names, paths, and gotchas live there.
Browser Automation
Pre-installed agent-browser CLI plus create_browser / delete_browser MCP tools for driving a remote Chromium with live-view streaming. Downloads have a non-obvious retrieval flow.
→ When the user wants to fill forms, scrape pages, click through a UI, or share a live view — see reference/browser.md.
Sandbox
create_sandbox MCP tools for isolated code execution and web-app preview URLs. Use it instead of polluting /workspace with throwaway runtimes.
→ When running untrusted code, exotic runtimes, or previewing a dev server — see reference/sandbox.md.
File Sharing
Shared folders mounted at /mnt/afs/<name> let agents hand off files without inlining them in call_agent payloads.
→ When passing files between agents, or when a target agent should read directly from disk — see reference/file-sharing.md.
Memory
Persistent, cross-session storage mounted under /mnt/memory/<store_id>/. Workspaces opt in by attaching memory stores; when attached, the entries are listed in the platform reminder. Use ordinary file tools — there are no dedicated memory tools.
→ Before reading or writing any file under /mnt/memory/, or when deciding what to persist — see reference/memory.md for the on-disk schema (frontmatter, type, index), recall workflow, and the conventions a background consolidation pass relies on.
Skills
skill_create_draft / skill_enter_edit / skill_publish lifecycle for user-authored skills (this __platform__ skill itself is platform-managed and not user-editable).
→ When the user wants to create, edit, or publish a skill — see reference/skills.md.
Builder Mode
workspace_*_propose / _apply (and <resource>_*_propose / _apply for the prompt library) let you propose configuration changes — system prompt, schedules, slash commands, skills, model/provider, prompt library — that the user approves via an in-chat card before they apply. Read tools (list_*, get_*) are always on whenever any Builder cap is enabled.
→ When the user describes a configuration change in conversation rather than going to the UI form — see reference/builder-mode.md for the propose/approve/apply contract, scope semantics, and cross-cutting rules (ownership, versioning, source XOR, timezone, reference safety).
{{#codex}}
Image Generation
Generated images land in /workspace/.home/.codex/generated_images/ and should be surfaced inline.
→ When generating an image for the user — see reference/image-generation.md.
{{/codex}}