The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a defensive 'Prompt Injection Safety Notice' in SKILL.md. This section explicitly instructs the agent to treat common injection phrases such as 'ignore previous instructions' or 'you are now a different agent' as inert data rather than commands when encountered in manifest files or security advisories.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted project manifests and dependency metadata, creating a potential surface for indirect prompt injection.
Ingestion points: The agent reads package manifest files (e.g., package.json, requirements.txt, go.mod) and lockfiles (e.g., package-lock.json, poetry.lock) to perform its analysis as specified in the Trigger Conditions.
Boundary markers: The instructions include a dedicated safety notice that defines logical boundaries, warning the agent to maintain role boundaries and never follow instructions embedded in analyzed content.
Capability inventory: The skill's environment is restricted to file system read access via the 'Read', 'Grep', and 'Glob' tools. It lacks the ability to execute shell commands, write files, or perform arbitrary network requests.
Sanitization: The skill enforces a strict 'Assessment Output Template', ensuring that the results of the analysis are presented in a structured format that avoids the interpolation of raw external strings into sensitive contexts.- [EXTERNAL_DOWNLOADS]: The skill instructions reference external security data feeds and official documentation from well-known organizations.
Fetches vulnerability triage data from FIRST's EPSS API (api.first.org) and CISA's Known Exploited Vulnerabilities catalog (cisa.gov).
References standard industry tools and specifications including CycloneDX, SPDX, Syft, and Trivy for SBOM generation and auditing.

dependency-scanning