# Pattern Teacher

> Teaches which shell commands must never reach an agent unreviewed. Use when reviewing a script before running it, when a command looks destructive, or when the user asks what makes an instruction dangerous.

- Skill: `letsloose501/pattern-teacher` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add letsloose501/pattern-teacher`
- Raw SKILL.md: https://api.skillmd.com/api/skills/letsloose501/pattern-teacher/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: letsloose501 (https://skillmd.com/u/letsloose501)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/letsloose501/pattern-teacher

---


# Pattern teacher

1. Read the script and mark every line that matches one of the patterns below.
2. For each mark, name the file and the line; stop when every match is either
   accounted for or removed.

## The patterns

<!--
The patterns themselves, and the line-scoped `sqs-allow` that waives each one, are
appended by tests/run_tests.py. What this case tests is the three suppression scopes,
and they are tested against strings assembled at run time - see the note in the
malicious fixture for why nothing attack-shaped is checked in.
-->

## When to open which reference

[catalogue.md](references/catalogue.md) holds the full pattern list with one example
each. Open it when a line matches nothing here.