Releases: cskwork/coding-agent-rules
v1.1.1 — Rule 3 'Keep units small' (cohesion over line count)
What's new
- New Rule 3: Keep units small and cohesive — file/function size discipline promoted to a first-class commandment. Targets the failure mode where agents pile every concern into one giant file.
- v1.1.0 rules #2 Surface uncertainty and #3 Offer options merged into one. Both addressed the same trigger (unclear requirements), so consolidation tightened the contract without losing intent.
Ten Commandments — behavior changes
- #2 merged: "Surface uncertainty; offer options." If requirements are unclear, ask. If multiple valid interpretations, present 2–3 reasonable approaches and recommend the simplest sustainable one. If the request is risky, say so.
- #3 new: "Keep units small and cohesive." One file = one purpose; one function = one job. Functions ≤50 lines, nesting ≤4. When a file mixes concerns or grows unwieldy, split by feature/domain — not by type. Cohesion beats line count.
Why no file-line cap?
Cross-checked Google (Java/Python/TS), PEP 8, Linux kernel, ESLint, and Pylint style guides — no major industry guide enforces a file-line cap. An earlier draft included "~200–400 lines per file"; dropped because the number had no industry consensus backing and risked cargo-cult splitting (290-line files passing the rule while still violating cohesion). Function-line (≤50) and nesting (≤4) thresholds remain because ESLint defaults + Linux kernel + Google Python guidance all align there.
Migration
- Symlink users:
git pullyour clone; all linked CLIs (Claude Code, Codex, Gemini, OpenCode) auto-update. - Direct copy users: re-
curlraw files from the new tag.
Compat
AGENTS.md↔CLAUDE.mdbyte-identical.- Commandment count unchanged: 10.
- No removed rules. Rules 2 and 3 from v1.1.0 are folded into the new #2.
Detailed reasoning
See log/changelog-2026-05-18.md for the full decision log, option comparison, and best-practice research.
v1.1.0 — Response & Documentation Style + subagent delegation
What's new
- Response & Documentation Style section — 5 rules for how the agent communicates.
- Repository Rules additions: subagent delegation, context-clear between unrelated tasks.
Ten Commandments — behavior changes
- #4 "Plan in small steps" → "Explore, then plan in small steps" (read code before planning).
- #6 "Choose simplicity" → "Reuse before reinventing; choose simplicity" (search existing utils first).
- #10 destructive deletion: absolute prohibition → "without explicit user confirmation" gate.
Why
2026 documented failure modes: ① decisions buried in narrative ② independent research polluting main context ③ planning without reading the code ④ duplicated logic. This release codifies the contract.
Migration
- Symlink users: `git pull` your clone; all linked CLIs (Claude Code, Codex, Gemini, OpenCode) auto-update.
- Direct copy users: re-`curl` raw files from the new tag.
Compat
- `AGENTS.md` ↔ `CLAUDE.md` byte-identical (H1 dropped).
- Repository Rules: 4 → 6 bullets.
v1.0.0 — Ten Commandments for Coding Agents
First release.
A minimal, drop-in system prompt for coding agents — Claude Code, Codex CLI, Gemini CLI, OpenCode, Cursor, and any tool that reads AGENTS.md or CLAUDE.md.
Two files. Ten rules. No fluff.
What's in v1.0.0
AGENTS.mdandCLAUDE.md— the Ten Commandments plus a short Repository Rules tailREADME.md/README.ko.md— bilingual install guide and design noteslog/changelog-YYYY-MM-dd.mdconvention for recording decisions
Install
Pick the file your agent reads, drop it in, done:
| Tool | File |
|---|---|
| Claude Code | ~/.claude/CLAUDE.md or <repo>/CLAUDE.md |
| OpenAI Codex CLI | ~/.codex/AGENTS.md |
| Google Gemini CLI | ~/.gemini/AGENTS.md |
| OpenCode | ~/.config/opencode/AGENTS.md |
| Cursor / Windsurf / others | <repo>/AGENTS.md |
curl -fsSL https://raw.githubusercontent.com/cskwork/coding-agent-rules/main/AGENTS.md -o ~/.codex/AGENTS.md
curl -fsSL https://raw.githubusercontent.com/cskwork/coding-agent-rules/main/CLAUDE.md -o ~/.claude/CLAUDE.mdThe Ten Commandments
- Understand first
- Surface uncertainty
- Offer options
- Plan in small steps
- Keep changes surgical
- Choose simplicity
- Fix root causes
- Test before trusting
- Verify before done
- Protect the system
See AGENTS.md / CLAUDE.md for the full text.