The Agent Skills Directory

[PROMPT_INJECTION]: The static detector flagged a safety instruction as a potential injection. The skill contains a specific 'Prompt Injection Safety Notice' that instructs the agent to ignore directives like 'ignore previous instructions' or role-play attempts if they appear within scanned manifest files. This is a defensive measure, not a malicious pattern.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from package manifests and security advisories.
Ingestion points: The skill uses Glob and Read tools to ingest content from files like package.json, requirements.txt, and various lockfiles.
Boundary markers: The skill includes a dedicated 'Prompt Injection Safety Notice' section providing explicit delimiters for behavior when encountering adversarial instructions in data.
Capability inventory: The skill is restricted to Read, Grep, and Glob tools. It produces analysis and reports but does not have the capability to execute code or modify the environment.
Sanitization: The instructions explicitly mandate the redaction of any sensitive values (credentials, tokens) found during analysis.
[EXTERNAL_DOWNLOADS]: The skill references several official security databases and tools.
Fetches vulnerability triage data from FIRST.org (EPSS) and CISA's official Known Exploited Vulnerabilities (KEV) catalog.
Recommends the use of industry-standard security tools from well-known organizations including Aqua Security (Trivy), Anchore (Syft), and the CycloneDX project.

dependency-scanning