Field notes on AI agents and developer tools. Tested on real machines, not summarized from press releases.https://agentkoji.com/enhttps://agentkoji.com/en/articles/coding-agent-comparison/https://agentkoji.com/en/articles/coding-agent-comparison/IDE, CLI or cloud is only the starting point. Five incidents we hit whichever we used, and the operating habits we kept.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/green-tests-prove-nothing/https://agentkoji.com/en/articles/green-tests-prove-nothing/We could disable any of eight guards and every test stayed green. How to catch decorative tests, the kind that multiply once agents write your tests.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/agent-done-verification/https://agentkoji.com/en/articles/agent-done-verification/An agent reported six commits and an open PR after making zero tool calls. A memory rule didn't stop it happening again. Hooks and git numbers did.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/ai-agent-vps-setup/https://agentkoji.com/en/articles/ai-agent-vps-setup/Getting an agent to run under systemd takes half an hour. The hard part is noticing when it is green but dead. Monitoring and safety lessons from an always-on agent.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/mcp-server-intro/https://agentkoji.com/en/articles/mcp-server-intro/A minimal, tested MCP server, plus how to defend your tools against agent inputs that look nothing like human typos.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-pricing/https://agentkoji.com/en/articles/claude-code-pricing/Pro is $20 a month, Max starts at $100. The June billing change was paused on its start day. What the official pages say as of October 2026.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-md-trim/https://agentkoji.com/en/articles/claude-md-trim/We cut one CLAUDE.md from 199 to 114 lines and another from 120 to 52. Each model generation turns old scaffolding into debt.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-cursor-codex/https://agentkoji.com/en/articles/claude-code-cursor-codex/Sixteen subagents died at once on a 429. We compare the three tools on instruction files, hooks and review output, not feature lists.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-hooks/https://agentkoji.com/en/articles/claude-code-hooks/Our git guard hook let +main and git -C slip through. What hook exit codes really mean, and how to fix a hook by measuring it.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/subagent-failures/https://agentkoji.com/en/articles/subagent-failures/A git worktree doesn't contain your gitignored verify script. All three parallel lanes hit that. Guards we put in prompts, and how we check results.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/agent-permission-design/https://agentkoji.com/en/articles/agent-permission-design/Let an agent read production but never write. We built a read-only SQL wrapper and checked how deny rules and hooks are evaluated.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/bot-monitoring-invariants/https://agentkoji.com/en/articles/bot-monitoring-invariants/In AI agent monitoring, the job failed every time, yet the heartbeat stayed green. Where you send the beat decides how your monitor lies. Holes beyond the four rules, reproduced.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/ai-review-rounds/https://agentkoji.com/en/articles/ai-review-rounds/A design plan needed four review passes before it got a GO. How to tell an endless loop from one you must not cut short, plus review CLI traps.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/local-llm-claude-code/https://agentkoji.com/en/articles/local-llm-claude-code/Even a 0.5B model took about 47 seconds just to read 16k tokens on CPU. The slowness is not generation; it is how much you send every turn.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/agent-memory-design/https://agentkoji.com/en/articles/agent-memory-design/Claude Code memory: our MEMORY.md index and a session-start handover file that grew to 247KB. How we write indexes and spot rotting notes.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-secrets/https://agentkoji.com/en/articles/claude-code-secrets/Claude Code API key management needs more than a deny rule: grep and Python can still read .env. We reproduced the leak paths with a dummy key.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-env-vars/https://agentkoji.com/en/articles/claude-code-env-vars/Why a Claude Code environment variable you set doesn't take effect: settings env vs shell export vs .env, and when each is read. Reproduced with dummy values.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-ci-costs/https://agentkoji.com/en/articles/claude-code-ci-costs/Claude Code GitHub Actions costs come from two places: API tokens and runner minutes. A worked estimate, a cache trap, and how to stop runaway jobs.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/same-fix-three-times/https://agentkoji.com/en/articles/same-fix-three-times/When Claude Code keeps making the same mistake: stop after two failed fixes, check what's really wrong, and write recurrence rules into CLAUDE.md.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/doubt-ai-root-cause/https://agentkoji.com/en/articles/doubt-ai-root-cause/Claude Code's stated root cause is sometimes wrong. Check controls, environment, artifact freshness and the measurement setup, in that order, before you dig.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/local-green-ci-red/https://agentkoji.com/en/articles/local-green-ci-red/Works locally but fails in CI? Check differences in order: the commit CI actually ran, the shell and commands, dependency resolution, then architecture.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/ai-design-rebuttal/https://agentkoji.com/en/articles/ai-design-rebuttal/AI design review is cheaper before any code exists. Ask a different model for the strongest rebuttal, check it against settled decisions, then revise.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/ai-deletion-refactor/https://agentkoji.com/en/articles/ai-deletion-refactor/When a Claude Code refactoring deletes thousands of lines, remove callers first. How to map dependency direction and how to slice commits, reproduced locally.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/llm-pick-extract/https://agentkoji.com/en/articles/llm-pick-extract/LLM structured outputs fix the shape, not the truth. Don't let the model write prose: have it pick from options and extract quotes, then verify them by machine.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/deploy-migration-accidents/https://agentkoji.com/en/articles/deploy-migration-accidents/Claude Code deploy accidents that hurt most aren't dramatic deletions but quiet ones: success messages that did nothing, tests that ran in production. Fixes included.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/claude-code-getting-started/https://agentkoji.com/en/articles/claude-code-getting-started/How to use Claude Code from install to your first question and first change, following the official steps, run locally, with the snags we hit along the way.Fri, 16 Oct 2026 00:00:00 GMThttps://agentkoji.com/en/articles/mcp-server-choose/https://agentkoji.com/en/articles/mcp-server-choose/Before hunting for recommended MCP servers, have criteria: who maintains it, transport, scope, output size, and how to narrow its permissions once installed.Fri, 16 Oct 2026 00:00:00 GMT