The Agent Skills Directory

[SAFE]: The automated detection of prompt injection patterns in skill.md is a false positive. The phrase "ignore previous instructions" appears in the security documentation section (§10) to explain how the skill handles malicious payloads found in external GitHub data, rather than being an injection attempt itself.
[EXTERNAL_DOWNLOADS]: The skill includes a setup hint in §8 recommending the installation of the official GitHub MCP server (@modelcontextprotocol/server-github) via npx. This is an informative reference to a well-known, trusted service used to facilitate the skill's primary function.
[SAFE]: The skill implements strong defenses against indirect prompt injection (Category 8). It specifies that untrusted data from GitHub (issues, PRs, files) must be rendered inside Markdown blockquotes to isolate it from the agent's instructions (§5 and §6). Furthermore, it strictly limits its own capabilities to a whitelist of read-only operations, refusing any write-side intents (§1 and §3).

mcp-github-bridge