2.6.0 Changed How Hooks Wait. A Triage Order for Turns That Never Finish
When an agent turn never finishes, the cause is often a hook you wrote rather than the model. Here is what changed in IDE 2.6.0, an audit script that finds hooks capable of stalling a turn, and how to give a hook its own cutoff.
Guarding the Turn Boundary in Non-Interactive Runs: A Command Table and a Fail-Closed Gate
A scheduled job meant only to record usage was calling the model on every run. Here is how I split commands into ones that start a turn and ones that do not, built a gate that refuses anything unclassified, and measured it against real scripts.
Who Approved the Right Side of &&? Splitting Shell Commands Before Matching Allow Rules
The approval dialog showed part of what actually ran. Here is a harness that splits compound shell commands without breaking quotes or command substitution, matches allow rules per segment, and the numbers from running it over 45 real commands.
When an Unresponsive MCP Server Freezes Your Agent: Separate Timeouts for Connect, List, and Call
Antigravity CLI 1.1.3 closed the case where an unresponsive MCP server stalls an agent forever, by adding timeouts to connect, list-tools, and call-tool. This walks through why the three boundaries fail differently, and builds a defensive wrapper with a circuit breaker and failure-only notifications, backed by working code and a week of overnight runs.
The One File That Stops Startup: Guarding config.json Integrity Before Scheduled Runs
Antigravity 2.3.1 fixed a bug where an empty or corrupt config.json blocked startup. The symptom was fixed; the causes of corruption still live in your environment. Here is a three-layer guard — validate, snapshot, restore — that runs before a scheduled job, with working code and notes from running it nightly.
Turning Silent Auto-Approvals into Allow Rules, One Soft-Deny at a Time
In Antigravity CLI 1.1.3, headless -p stops silently auto-approving confirmation-required tools and instead soft-denies them, printing the required allow-rule name to stderr. This piece uses that output as a discovery source to build least privilege from an empty allow set upward, with a working harness and real numbers from a personal automation.
One Space in a Path, and Nine Commands Reported Success While Counting the Wrong Place
A single space in a workspace name sends agent-written commands somewhere else, quietly. Measurements across eleven unquoted-path forms, and the entry-point script that closes the boundary in one cd.
When Your Agent Commits a .bak File: Why Fix-Tool Artifacts End Up in Git
Backup files like .bak and .orig slip into commits after an agent runs a --fix tool. Here are the reproduction conditions, the real root cause, and three fixes: narrowing the staged scope, wrapping the fixer, and adding a pre-commit extension gate.
When an Unattended Run Finds Its Working Folder Owned by Someone Else
A schedule that had been green for weeks failed one morning because it could not write .git/config. The cause was ownership drift on a reused working folder. Here is how to tell apart the read-but-not-write trap, probe writability for real, and fall back to a folder you can always write.
The Gate That Stops Visual Damage Before You Hand Bulk Image Optimization to an Agent
When an agent bulk re-encoded a few hundred wallpaper assets, a handful came back with dulled color. Size-reduction alone cannot catch that. Here is how to design a gate that stops bad conversions before merge using three axes — SSIM, ΔE, and file size — with a checker that runs on Pillow and scikit-image.
What to Delegate to an Antigravity Agent and What to Keep by Hand, After Two Weeks
After two weeks of handing my daily solo-dev tasks to Antigravity agents, a clear line emerged between the work I was glad to delegate and the work I had to pull back. A retrospective with the operational log.
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.