# Loader

> A general-purpose helper whose module-level TEMPLATE constant (assembled from two chunked/part files) shares a bare name with an unrelated function's own local eval() target -- name shadowing, not data flow. Leg 3 corroborates on its own; the taint must still not cross scope.

- Skill: `gl0di/loader-4` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add gl0di/loader-4`
- Raw SKILL.md: https://api.skillmd.com/api/skills/gl0di/loader-4/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: gl0di (https://skillmd.com/u/gl0di)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/gl0di/loader-4

---


# Loader Skill (clean name-shadow regression)

`_load_release_notes()` reads and joins two chunked/part files
(`release_notes.part1.txt`, `release_notes.part2.txt`) into a module-level `TEMPLATE`
constant -- inert text, never executed. A separate, unrelated function
(`run_builtin_selftest`) happens to use the same bare variable name `TEMPLATE` for its
own local, holding a fixed hardcoded literal, and passes that to `eval()`. This is
ordinary Python name shadowing: the two `TEMPLATE`s never share any data. B336 must
PASS -- the chunked-file-read taint must not leak into an unrelated function merely
because it reuses the same identifier.