Decky Loader AI Development Policy — Claude steering notes for decky-proton-pulse

gistfile0.txt

# Decky Loader — AI Development Policy (Claude Steering)

According to the current consensus among the Decky Loader maintainers and community guidelines,
using AI to develop plugins is acceptable but subject to strict scrutiny during the PR process.

## 1. Core Policy: "Supplementary, Not Central"

- **Assistive Use (OK):** Using AI for boilerplate, API calls, database strings — standard modern workflow.
- **Generative Creators (Discouraged):** AI as the effective "author" is a red flag — subtle bugs, security holes, unoptimized code the submitter can't explain.
- **AI Wrappers (Likely Denied):** Plugins that are simply wrappers for ChatGPT/Claude etc. are discouraged. Prefer plugins providing functional utility to the Steam Deck interface.

## 2. PR Process is the Deciding Factor

- **Human Accountability:** You must be able to explain, defend, and debug every line submitted.
- **Detailed Criticism:** Expect longer-than-usual review. "Hallucinated" code or patterns that don't align with Decky's React/Python bridge will likely be rejected.

## 3. Non-Negotiable Standards

- **Security:** Decky runs with significant permissions. Improper fetch calls or shell command handling = immediate rejection.
- **Performance:** Avoid background battery drain patterns common in AI-generated code.
- **Licensing:** No GPL-conflicting snippets. Must include a valid LICENSE or LICENSE.md file.

## Summary Checklist

- [ ] Disclose AI involvement in the PR description if significant
- [ ] Manually verify all React components and Python backend logic
- [ ] No unsafe shell command handling or improper fetch calls
- [ ] No background battery drain patterns
- [ ] Valid LICENSE file present (GPL-3.0 preferred)
- [ ] Plugin solves a real SteamOS/Bazzite user problem — not a tech demo

## The Bottom Line

Treat the AI as a junior assistant. You are the lead engineer who signs off on safety and quality.