All Articles
You Can Measure a Request Before You Send It — Sizing Agent Tasks by Working Backward from Rework Rate
When an Antigravity agent returns code that misses the mark, the cause is rarely the wording of the prompt. It is the size of the task. Here is a Python scorer that grades a request before you send it, plus what happened when I scored 80 past requests against their actual rework outcomes.
When Lighthouse Is Green but Search Console's Core Web Vitals Are Red — Field Notes on Naming the Slow Interaction with Real-User Data
Lighthouse scores in the 90s, yet field Core Web Vitals won't budge. Here is how I closed the lab-vs-field gap with real-user monitoring (RUM), named the exact interaction driving a slow INP, and fixed it with Antigravity.
When Antivirus Blocks Antigravity's Launch, Verify the Signature Before You Allow It
When the Antigravity desktop app won't launch because antivirus stepped in, here's how to tell a false positive from a real threat. Verify the code signature on macOS Gatekeeper and Windows Defender first, then allow-list with the narrowest possible scope.
When Your Agent's Commits Pick Up Junk Files: Fixing It With Staging Scope and a Message Convention
Agents tend to run git add -A, sweeping .bak files and caches into your history, and leave a one-word message. Here is how a staging allowlist, a preflight, and a fill-in message template stop it.
Where Context Should Live — Routing It Across Persistent, Per-Task, and In-the-Moment Layers
Antigravity 2.0's /btw lets you inject context mid-task. But unless you separate what belongs in a persistent rule, a task instruction, and a passing aside, your agent's accuracy actually drops. Here is a design for three context layers.
Don't Let Your Automation Lean on AI Ultra's 5x Ceiling
The $100/month AI Ultra plan raises Antigravity's usage limits to 5x AI Pro. But if you architect automation around that ceiling, it collapses the moment you drop back a tier. Here is a limit-independent degradation design, with the real pain points.
Confirm Your Unattended Runs Survived a Point Release in 30 Seconds
As Antigravity ships point releases at a fast clip, how do you stop scheduled runs from quietly breaking after an update? Here is a lightweight smoke test that detects regressions against a golden output and rolls back automatically when it diverges.
Hearing the Audio an Agent Made, Right Inside the Conversation
A recent Antigravity point release added inline audio rendering in the conversation view. Here is how playing agent-made audio in place changes the way I audition sound assets for my apps.
Before Desktop and CLI Drift Apart: Put Agent Steps in One Versioned File
As Antigravity 2.0 multiplies entry points across desktop, CLI, and SDK, the instructions for the same task slowly diverge per surface. As an indie developer running several sites on autopilot, I lay out a design that consolidates the steps into a single versioned file each surface merely reads.
Strip Secrets Out of the Agent Logs You Keep: Designing a Redaction Layer
Once you start keeping logs from unattended agents, a token or API key eventually lands in them in plaintext. Rotating the key doesn't unmake the leaked log. This designs a redaction layer that reliably drops secrets right before the write, going beyond regex to register known secrets and mask them for certain, with working Python and field notes.
Tracing What a Long Agent Run Actually Did: Review That Starts From In-Conversation Search
How to use the in-conversation search added in Antigravity v2.1.4 as the starting point for reviewing long agent runs. Choosing search terms, the decision points to inspect, and reconciling with background-agent logs, with concrete steps.
Keeping Scheduled Runs Reproducible: Pinning the Antigravity CLI Version to Tame Behavior Drift
The Go-based Antigravity CLI is now available to everyone, and updates are landing at a quick pace. When a CLI baked into your automation upgrades underneath you, a single morning's job can behave differently. Here is how I keep things reproducible — pinning the binary, recording its identity in each run's log, and rolling upgrades forward one job at a time — drawn from running four sites on an overnight schedule.