The Agent Skills Directory

[External Content Retrieval]: The skill fetches repository listings and markdown files from remote registries (GitHub or external URLs) to index and preview skills. While this is the primary purpose of the skill, retrieving data from arbitrary user-provided URLs is a security consideration that requires oversight.
[Indirect Prompt Injection Surface]: By fetching and displaying SKILL.md files from community-contributed registries, the skill creates a surface for indirect prompt injection. A malicious registry could provide content designed to influence the agent's behavior while the user is browsing or previewing matches.
Ingestion points: Step 1 (Fetch registry indexes) and Step 4 (Preview) in SKILL.md fetch content from remote URLs.
Boundary markers: Not explicitly defined in the fetching logic.
Capability inventory: The skill can read/write to local configuration files and perform network requests.
Sanitization: No specific sanitization or escaping of external markdown content is mentioned prior to presentation.
[Configuration Modification]: The skill is designed to update the CLAUDE.md file in the user's plugin configuration directory to add new registries. The instructions include a requirement for user confirmation before performing this action, which helps mitigate the risk of unauthorized persistence or configuration changes.

registry-browser