DevAgentRadar → Google Antigravity

Goal: turn this release signal into a safe pilot plan for my stack.

Context

Assistant: Google Antigravity Proposal: Grant URL-fetch access and retune -p timeouts after Antigravity 1.1.28 Summary: Antigravity 1.1.28 asks before fetching URLs, returns partial output with a stderr warning when --print-timeout expires, and marks fatal headless errors with error:. Update URL grants, CI exit handling, plugin reinstalls, and model-resolution log review so unattended runs do not stall or mis-score. Primary source: https://github.com/google-antigravity/antigravity-cli/releases/tag/1.1.28

Why it matters

1.1.28 changes defaults unattended jobs still assume. URL fetches are no longer always allowed; the agent prompts Allow access to this URL? unless access was already granted, so -p jobs that read the web will wait on approval. --print-timeout mid-turn now returns partial stdout, warns on stderr, and exits 0 (Ctrl+C still non-zero). Fatal headless errors use a stable error: stderr marker plus a truncation note, so parsers that treated timeout or empty output as failure will false-pass or miss failures. Tool prompts now name the action (Run this command?, Allow access to this URL?, Allow calling this tool?) and may include a Reason: line. Plugin install replaces the managed directory exactly and refuses installing from the installed copy; plugin MCP servers resolve cwd against the plugin directory. Subagent MCP-init races, stuck-running subagents, terminal-command memory leaks, and headless plan-review stalls are fixed. The CLI log records when a requested model resolves to another (alias, --effort, deprecated replacement).

Suggested actions

1. Upgrade Google Antigravity to 1.1.28. Before the next unattended -p run that reads the web, grant URL-fetch access so the job cannot block on the new Allow access to this URL? prompt (URL fetches are no longer always allowed).
2. Update headless -p CI for --print-timeout: treat expiry as a successful partial result (stdout plus stderr warning, exit 0). Fail the job only when stderr contains the stable error: marker or the process was interrupted; record the truncation note when it appears.
3. Reinstall each plugin from its source so 1.1.28 replaces the managed directory exactly; do not install a plugin from its own installed directory. Restart so plugin MCP servers resolve relative working directories against the plugin directory.
4. After pinning a model or passing --effort, open the CLI log and confirm the recorded resolution (alias, --effort variant, or deprecated-saved-model replacement) is the model that actually ran.
5. Keep long sessions that run many terminal commands on 1.1.28 instead of restarting to reclaim memory. If a subagent still looks running after finish or when-idle messages sit queued, stay on 1.1.28 (stuck-running and MCP-init races are fixed) rather than killing the session.

Config surfaces this release may change

• background and headless runs (high confidence) — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/google-antigravity-1-1-28-grant-url-fetch-access-and-retune-p-timeouts-a and mark Applied, Skipped, or Failed. Proposal id: a147aeb6-0c96-41ce-9456-b52101c35530

Please:

1. Map the change to concrete pilot steps
2. Flag security / permission implications
3. Keep the pilot reversible