SEC416 / MD-CLAUDE-PIP-INSTALL

SEC416 flags AI-native markdown when the same document explicitly says to prefer uv over pip, but still models Claude package installation with bare pip install.

Why It Matters

If a project teaches agents to prefer uv, transcript examples that still show Claude using bare pip install create contradictory guidance. These examples are easy to copy into real agent workflows and make package-management behavior less consistent and less reproducible.

This rule now lives in the explicit guidance lane because it measures consistency inside AI-facing documentation rather than a standalone exploit primitive. The signal is actionable, but the first release keeps it outside default security review lanes while broader ecosystem usefulness is measured.

Trigger Shape

The rule triggers only when all of these are true:

• the file is an AI-native markdown surface such as CLAUDE.md, AGENTS.md, SKILL.md, Cursor rules, or plugin agent docs
• the document contains an explicit uv-over-pip preference marker such as use uv not pip or Always use \uv` instead of `pip``
• a prose, blockquote, or fenced transcript region contains a Claude: line with bare pip install or python -m pip install
• the matched Claude: line does not already use uv pip install

Clean Cases

These stay clean:

• Claude: uv pip install pytest
• bare pip install examples without any explicit uv not pip preference context
• generic package installation prose that is not framed as a Claude: transcript

Example Trigger

- Always use `uv` instead of `pip` for Python packages

Claude: pip install pytest

Safer Example

- Always use `uv` instead of `pip` for Python packages

Claude: uv pip install pytest

How To Fix

Replace bare pip install Claude transcript examples with uv pip install, or clearly mark them as intentionally incorrect pre-correction behavior if the document is demonstrating a before-and-after learning flow.