intuitive-initlisted
Install: claude install-skill MiaoDX/intuitive-flow
# Intuitive Init
Set up repo-local AI agent guidance without turning shared defaults into a
symlinked source of truth. Shared skills should travel across projects;
`AGENTS.md` and `CLAUDE.md` should preserve the local repo's commands,
constraints, workflow choices, and hard-won mistakes.
In Claude Code and Codex work, a repo harness is the infrastructure that lets an
agent enter an existing project without guessing: root and nested instruction
files, reusable skills, hooks, MCP configuration, target-repo LSP setup, and
local verification commands. `$intuitive-init` builds or refreshes that harness
from repo evidence, official tool guidance, and init-style suggestions.
Default posture: keep root agent files aggressively small. Correct but lengthy
procedures should usually move out of `AGENTS.md` and `CLAUDE.md` into
`docs/agents/**`, reusable skills, or scripts, with the root files keeping only
the rule, trigger, and pointer.
Use these size signals:
- Target: each root agent file is short enough to skim before work starts,
usually under 120 lines.
- Warning: over 180 lines means audit should report bloat and propose deletes or
extraction.
- Strong cleanup signal: over 250 lines, duplicated sections between
`AGENTS.md` and `CLAUDE.md`, or long numbered procedures in root files.
These are signals, not hard limits. Keep a longer root file only when the
content is a critical safety rule that agents must see before any other read.
## Official Reference Sources
Prefer