# Core

> Required reading. Read this first at the start of every session, before anything else. It holds the minimal rules that apply to all work, plus the routing table that says which other skills to read. Other skills are read only when the routing table or their own description says so.

- Skill: `flll/core` (Agent Skill)
- Install (CLI): `npx skillmds@latest add flll/core`
- Raw SKILL.md: https://api.skillmd.com/api/skills/flll/core/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: flll (https://skillmd.com/u/flll)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/flll/core

---


# core — the minimal rules for every session

<!-- Fill in every <...> with your own words. Keep this file at 5-10 rules.
     When it grows past that, move the overflow into a new skill and add
     one row to the routing table below. -->

1. Respond in <your language>. (Example: respond in English, commit messages
   in English too.)
2. When reporting work as done, include the commands you ran and their real
   output. Never say "this should work" without having run it.
3. Before any task that touches multiple files or deletes anything: restate
   the request in 3 lines, confirm the goal / definition of done / allowed
   scope, and get agreement before writing.
4. When the user corrects you about the same thing for the second time, read
   `skill-coach` and propose turning the correction into a skill.
5. <your project-specific rule> (Example: never push directly to main.)

## Routing table

If a request matches a row below, read that skill's SKILL.md before starting
the work. Every new skill gets one row here.

| When the request involves... | Read |
|---|---|
| a repeated correction, or creating/improving/pruning a skill | skill-coach |
| (example: commit / PR / branching) | (example: git-rules) |