Every rule, with its source.

Bidirectional control characters

Comments hidden in rendered views

Hidden comment that reads like an instruction to the agent

Inline style or the hidden attribute hides text

Invisible Unicode characters

Decodes base64 or hex and runs the result, or hides a command in an encoding

Project hook or settings command fetches remote code, calls the network or redirects the API

A key or token is written in plain text in an MCP server or settings env block

MCP server runs an unpinned package or image, a downloaded script, or connects over plain http

The folder has a .env file but no permissions.deny rule keeps Claude Code out of it

Allow rule or trust flag skips the prompt for any shell command, network tool or interpreter

Settings turn on bypassPermissions or skip its confirmation

Downloads code and runs it (curl or wget piped to a shell, and similar)

Turns off the agent's permission checks

A script's comment or string gives the agent covert instructions

Invisible or bidirectional characters in a skill script

Reads a secret and sends data over the network in the same script or block

allowed-tools grants unscoped Bash, all tools, any interpreter or unscoped WebFetch

Skill argument is pasted into a !command that runs when the skill loads

Executable binary inside a skill folder

Skill script makes network calls

HTML comment never closes

File path contains control, bidi or invisible characters

@import names a file that does not exist

Relative link or anchor points to nothing

compatibility is longer than 500 characters

Skill description is longer than 1,024 characters

File is not UTF-8

A frontmatter field has the wrong type

File with line breaks does not end with one

A byte order mark comes before the frontmatter

Frontmatter does not start on line 1

Frontmatter is valid YAML but not a key: value mapping

Frontmatter is too large to parse safely

Frontmatter opens with --- and never closes

An unquoted value is cut short at " #"

Frontmatter is not valid YAML

@import reaches outside the project

File contains invalid UTF-8 bytes

Link points outside the project and was not checked

Memory file type is not user, feedback, project or reference

Copilot instructions file has neither applyTo nor description

Required description field is missing

A file that needs frontmatter has none

Required name field is missing

File mixes CRLF, LF or CR line endings

Skill name does not match its folder name

Skill name is not lowercase letters, digits and single hyphens

Skill name contains anthropic or claude

Skill name is longer than 64 characters

Cursor rule has no alwaysApply, globs or description

File contains NUL bytes

Memory topic file is not linked from MEMORY.md

Lines end in spaces or tabs that change nothing when rendered

A frontmatter key the reading tool ignores

Skill name or description contains an XML tag

SKILL.md body is over about 5,000 tokens

AGENTS.md files from the root down exceed Codex's 32 KiB

Over 5,000 tokens load every session from this file

description plus whentouse is cut in the skill listing

The same rule is in two files one tool loads every session

The same rule appears twice in one file

File is longer than the vendor recommends or reads

Claude Code reads CLAUDE.md here and skips AGENTS.md

AGENTS.override.md, AGENTS.local.md or .agents/*.md is read by some agents only, or by none

A CLAUDE.local.md switches off the AGENTS.md fallback

Two lines in one file say opposite things

Two files loaded for the same agent, or mirrored for different agents, contradict

Mirrored instruction files share most rules but not all

Agent config file is not valid JSON

Inline suppression of a security rule

asitis-disable names a rule that does not exist, or no rule at all

asitis-disable comment suppresses nothing