← ClaudeAtlas

antigravity-rescuelisted

Use when stuck, wanting a second implementation or diagnosis pass, or handing a substantial task to Google's Antigravity CLI (agy) through a real subprocess -- a genuinely different agent harness, not just another model. Not relevant when already running under Antigravity CLI itself.
patrickserrano/lacquer · ★ 1 · AI & Automation · score 67
Install: claude install-skill patrickserrano/lacquer
# Antigravity Rescue A thin forwarding wrapper around Google's Antigravity CLI (`agy`). Your only job is to forward the task and return its output. Do not inspect the repo, read files, draft your own solution, or do any independent analysis — that defeats the purpose of getting a second, independent opinion. ## Security: `agy -p` has no enforced sandboxing Verified live (three separate tests, all on `agy` 1.0.6): in headless print mode, `agy` executed a shell command, deleted a file inside the declared `--add-dir` workspace, and deleted a file **outside** it — all instantly, with zero confirmation, whether or not `--dangerously-skip-permissions` was set. That flag appears to only affect the interactive TUI's confirmation dialogs; headless mode has no technical permission boundary at all. `--add-dir` is a workspace *hint* (it decides where ambiguous file operations default to when unspecified) — it is **not** an access restriction, and does not confine `agy` to that directory. Practical consequence: `--read-only` below is **prompt-compliance only** — the model choosing to honor an instruction, not anything CLI-enforced. There is no equivalent of Claude Code's `--permission-mode plan` here. Forward a task expecting the same trust you'd extend to giving `agy` unrestricted filesystem access under your own user account, regardless of which flag you pass — because that's what you're actually doing. ## When to use - Proactively, when stuck: a recurring error, an approach that