← ClaudeAtlas

intenticsolid

What Intentic is and how this sandbox works. Load it BEFORE answering any question about the product (a panel, setting or card; connecting, configuring, extending or debugging the sandbox), before saying Intentic cannot do something, and whenever the sandbox misbehaves (a failed turn, a broken editor).
intentic/intentic · ★ 46 · Code & Development · score 82
Install: claude install-skill intentic/intentic
# Intentic: the product you are running inside Intentic is a per-workspace AI-agent development environment. This container is its **sandbox**: a daemon (`@intentic/sandbox`, image `ghcr.io/intentic/sandbox`, installed under `/opt/sandbox`) that serves one workspace, `/work`, on the owner's own host or a machine they chose. The owner drives it from a browser **editor**; a hosted **platform** handles sign-in, capabilities and the hosted plan. You are one **agent turn** the daemon is running. What the daemon does around you: - **Conversations are agents on worktrees.** Each conversation works on its own git branch (`agent/<conversation-id>`), checked out in a worktree that is mounted over `/work` for the turn. When a turn ends cleanly its delta is **landed**: applied to the owner's main tree as UNCOMMITTED changes, so their own commit is the review boundary. That is why you commit only when asked. A patch that will not apply refuses the whole land and raises a conflict card naming the paths; every worktree keeps everything. Untracked files shaped like scratch (a new hidden folder, logs and dumps, a checkout of its own) never ride a land: they stay in the conversation's copy, listed on its review, until the owner includes or deletes them. - **Runtimes.** A turn runs on Claude Code (this loop), native Codex, OpenCode (Grok, Gemini), Pi, Cursor or an ACP agent, chosen per conversation. Which model actually ran is recorded (`mcp__diagnostics__turns`). - **Capabilit