The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scans have confirmed the presence of 6 malicious URLs associated with the domain najaminstitute.com within the skill files. These URLs (including paths like /zakat and /about-us) are currently blacklisted and pose a severe risk of redirection to malicious content or phishing attempts.
[PROMPT_INJECTION]: The application implements a data-driven narrative rendering engine in SourceLinkedText.tsx that parses custom <cite> tags from evaluation data. This architecture creates an indirect prompt injection surface where the agent could ingest and obey malicious instructions embedded in the charity evaluation JSON files.
Ingestion points: evaluation narrative, headline, and summary fields in data/charities/charity-*.json files.
Boundary markers: Absent in the parsing logic; instructions in data are not delimited from the system context.
Capability inventory: The application has access to Firebase authentication, Firestore write operations (reporting issues and logging donations), and external URL redirection.
Sanitization: Although dompurify is included in the project's package.json, it is not utilized in the manual regex-based parsing logic of the evaluation text components.

frontend-design