AI - Loop Notes

Antigravity CLI Remote Control: Setup, Fixes, Approvals

November 3, 2026October 25, 2026 by jjeongil

Three-step flow for agy: command not found. The agy binary is installed in ~/.local/bin, agy --version fails with command not found, and adding ~/.local/bin to PATH makes agy work

Antigravity CLI Remote Control lets you follow a session and answer approvals from your phone. Session vs daemon mode, the agy PATH fix, and limits I hit.

Cursor Rules Not Working? Check the Glob and Description

November 3, 2026October 22, 2026 by jjeongil

Example Cursor rules structure: a few Always Apply rules on top, with area rules below split by file pattern using Apply to Specific Files

Cursor rules not applied? Each rule type has a condition. Mine failed on a too-narrow glob and a vague description. How to check and fix both.

Karpathy CLAUDE.md Tested: Does It Change Claude Code?

November 2, 2026October 21, 2026 by jjeongil

Stacked bar chart from a Karpathy CLAUDE.md test: with the file, Claude Code asked first in 3 of 6 runs and stated its reading in the other 3; without it, all 6 runs picked a reading silently

I ran Karpathy’s CLAUDE.md through 60 Claude Code runs: it made the agent ask first on a vague task 3 of 6 times, but barely changed bug fixes or tests.

x402 Agent Spending Limits: The SDK Cap and the Rest

November 2, 2026October 21, 2026 by jjeongil

Example x402 client payment flow: request, 402 response, check payment requirements, then pay and resend once if within the limit, and stop if over the limit or if it still fails

The x402 spec leaves agent budgets to you. Since SDK 2.23.0, spendControls caps each payment at $1 by default. Per-task and daily limits are still yours.

Claude Code Not Following CLAUDE.md? I Tested It 57 Times

November 2, 2026October 19, 2026 by jjeongil

Bar chart: with each CLAUDE.md rule, 5 of 5 Claude Code runs did what the rule asked; without the rule, 0 of 3 did for six rules and 1 of 3 for the legacy folder rule

I ran Claude Code 57 times against 8 kinds of CLAUDE.md rules with no-rule controls, and 45 of 45 held, so here’s what I’d check when yours get ignored.

Claude Code Custom Commands to Skills: Why /project: Fails

November 1, 2026October 18, 2026 by jjeongil

Diagram of a single command file in .claude/commands and a skill folder with SKILL.md in .claude/skills, both called as /deploy

/project: fails because Claude Code merged custom commands into skills. What changed, what broke when I moved mine, and the description fix that helped.

AI Agent Ran git commit –no-verify? Make CI the Real Gate

October 31, 2026October 17, 2026 by jjeongil

Flowchart: an AI agent's commit fails the pre-commit hook, is re-committed with git commit --no-verify, then fails lint in GitHub Actions CI

git commit –no-verify skips pre-commit and commit-msg hooks, so no hook can stop it. My AI agent used it; CI as a required check caught the lint error.

Claude Outage Fallback: Request Failed, Files Still Changed

November 1, 2026October 15, 2026 by jjeongil

Recreated Claude status incident record for September 29, 2026, showing the title Elevated errors and the 14:00–14:59 UTC impact window, with update rows hidden

Claude down mid-task? A failed agent response can still leave edited files. When switching to Codex is safe, what –fallback-model covers, how to resume.

Antigravity CLI Always Proceed Still Asks? Permissions Tested

October 30, 2026October 12, 2026 by jjeongil

Diagram showing Artifact Review as a separate flow, while a command passes through the Tool Permission mode and fine-grained rules, with an outside-the-sandbox boundary, to allow, ask, or deny

Antigravity CLI still asking with Always Proceed on? Check ask rules and Artifact Review first. 5 runs per mode: 30% faster, but review caught 4 actions.

RAG Retrieval vs Generation Failure: How I Tell Them Apart

October 31, 2026October 11, 2026 by jjeongil

Diagnostic flow from expected evidence to Top-k, final context, and answer, showing which RAG failure label applies at each stage

Wrong RAG answer? If the right chunk never reached Top-k, it’s retrieval. If it was in the final context, it’s generation. Logs from my ~50-question POC.