The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a 'Prompt Injection Safety Notice' that instructs the agent to ignore and treat any instructions found within analyzed data (such as manifest files or advisories) as data rather than commands. This includes specific mentions of phrases like 'ignore previous instructions' and 'you are now a different agent'. These are defensive measures intended to prevent the agent from being subverted by malicious content in the files it scans.
[EXTERNAL_DOWNLOADS]: The skill references and instructs the agent to utilize external vulnerability and threat intelligence feeds from well-known security services. This includes the CISA Known Exploited Vulnerabilities (KEV) catalog and the FIRST Exploit Prediction Scoring System (EPSS) API. These references are to official, established security resources for the purpose of data enrichment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (software manifests, lockfiles, and advisory feeds), creating an attack surface for indirect prompt injection.
Ingestion points: Manifest files (package.json, requirements.txt, csproj, etc.), lockfiles, and external security advisory data.
Boundary markers: Present. The skill includes an explicit 'Prompt Injection Safety Notice' that provides guidance on maintaining role boundaries and ignoring instructions embedded in analyzed content.
Capability inventory: The skill's allowed tools are limited to Read, Grep, and Glob. There are no tools for arbitrary code execution (eval/exec) or writing to the network.
Sanitization: The instructions explicitly command the agent to redact or generically reference any sensitive values (credentials, API keys, tokens) found during the analysis process.

dependency-scanning