A complete AGENT.md: one dev's tuned global ruleset for coding agents
alexcovo_eth · x · 2026-09-06
Khazix0918 shares a fully rewritten AGENT.md global config for coding agents, ready to copy.
- Communication: user-facing narration in Simplified Chinese (code/commands in English), conclusion first, coherent paragraphs over reflexive bullet lists, no filler or unsolicited comparisons.
- Priority: current explicit user instruction > Skills, memory, defaults; project-level AGENTS.md only overrides globally within that project.
- Execution: keep pushing to the goal once work starts; do already-authorized work before asking questions so the user approves concrete, checkable results; push back on bad suggestions; never proactively add disclaimers or approval gates.
- Testing: no tests for trivial reversible changes; run proportionate tests, then don't expand coverage without new changes or failures.
- Tooling: prefer rg and batched reads; use a logged-in Chrome when no CLI exists; delegate to subagents only for genuinely independent workflows with clear inputs, outputs, and completion criteria, with the main agent verifying results.
More from coding & agent
- Grok bot asks permission to spend 590 CZK on Czech e-commerce, gets instant approval — jeff_weinstein · 2026-09-06
- DataFast MCP update lets AI agents answer questions about bot traffic — marclou · 2026-09-06
- Grok bot feels like GPT-3.5 in testing, lacks 'done receipts' — pudding protocol proposed — RachelVT42 · 2026-09-06
- Stripe openly solicits integrators for Link's agent wallet features — jeff_weinstein · 2026-09-06
- Stripe engineer explains how Link's next_action guides agents through payment permission steps — jeff_weinstein · 2026-09-06
- A 15-item stack checklist for letting coding agents run an entire company — aryanXmahajan · 2026-09-06