This will be available in the next version. When an agent starts, its attached skills (folder, project, agent and ticket skills) are now also written in the CLI's own skill format, in a folder only that session reads. With Claude Code they show up in /skills, work as slash commands and load when a task matches; Antigravity gets them the same way. Nothing is written into your repository or your own skill folders, and the set is refreshed at every launch and restart. A new per-agent setting, "Skills delivery" (agent editor, right under Default skills, also on mobile), lets you choose prompt text, native skills, or both (the default). Codex cannot load skills for a single session yet, so it keeps the prompt text. If this is not quite what you had in mind, tell us and we will adjust it.
Deliver a session's attached skills as native CLI skills, not only as prompt text
From bensDone 14 hours ago
AgentsRoom already resolves which skills each session gets (agent defaults, project and folder skill sets, ticket skills with merge/replace), but delivers them only as text in the agent's first message. The CLI itself never sees them as skills: they don't appear in its native skill list, can't be invoked as slash commands, and aren't auto-matched to tasks (for example, in Claude Code they are missing from /skills).
The existing per-skill export to a CLI folder is manual and writes to a shared project or home location, so every session sees every exported skill regardless of its task.
Proposal: at session launch, materialize exactly that session's resolved skill set in the CLI's native skill format (main file plus bundled files) into a location only that session reads (for example, the per-agent configuration directory AgentsRoom already manages). Refresh on restart/clear, write nothing into the repository, and never touch user-created skill folders. Add a per-agent or per-CLI option: text injection, native skills, or both.
Outcome: what you attach in AgentsRoom is exactly what the CLI's native skill picker and slash commands offer for that session. Related: a separate request for reusable "Toolkits" decoupled from persona and memory.
AgentsRoom diagnostics
- App version: 1.225.0
- OS: darwin 27.0.0 (arm64)
- Agent CLI: claude
- App language: system
Problem to solve
Attached skills reach agents only as prompt text, so the CLI's native skill list, slash commands and auto-matching never see them; manual export is shared across all sessions.
Proposed behavior
At launch, write the session's resolved skills in the CLI's native format to a session-private location; refresh on restart; never write to the repo or touch user folders; option for text, native, or both.
Who benefits
Users running several agents or tasks that each need a different skill set.
Acceptance criteria
What is attached in AgentsRoom equals what the CLI's native skill picker and slash commands offer for that session.