# Research Gate

> Use when writing or editing .py files. Enforces approved-source research before code writes. Trigger: any .py file creation or modification. This skill reads config/approved_sources.json and validates that research.md contains URLs from approved domains that actually resolve. Blocks writes without valid research.

- Skill: `0sxd/research-gate` (Agent Skill)
- Install (CLI): `npx skillmds@latest add 0sxd/research-gate`
- Raw SKILL.md: https://api.skillmd.com/api/skills/0sxd/research-gate/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: 0SxD (https://skillmd.com/u/0sxd)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/0sxd/research-gate

---


# Research Gate

## What it enforces

Before ANY .py file can be written or edited:
1. A `research.md` file must exist at the target project root
2. It must be fresh (modified within the configured window)
3. It must contain URLs from domains listed in `config/approved_sources.json` for the active project
4. URLs must actually resolve (not just domain-match — the link must be real)
5. Citations are logged to `.claude/research_citations.log` (append-only TSV)

## Approved sources

Sources are defined per-project in `config/approved_sources.json`. The agent NEVER modifies this file. Only Sage adds entries via `/sagex:approve-source` or `/sagex:edit-mode`.

Default approved domains: `arxiv.org`, `github.com`

Project-specific domains inherit from global + project-level entries.

## What happens on failure

- **No research.md**: BLOCK. Agent must research first.
- **Stale research.md**: BLOCK. Agent must refresh research.
- **No approved URLs**: BLOCK. Agent must find approved sources.
- **URLs from unapproved domains**: BLOCK. Denial message lists approved domains.
- **URL doesn't resolve**: BLOCK. Agent must verify the link works.

## Integration

This skill describes the behavior. The mechanical enforcement is in `hooks/validate_research.py` which fires as a PreToolUse hook on Write/Edit events. The skill exists so agents understand WHY they're being blocked and what to do about it.