The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data, including Dockerfiles, Kubernetes manifests, and Helm charts, which creates an attack surface for indirect prompt injection. An attacker could place malicious instructions within metadata fields such as annotations, labels, or comments to influence the agent's behavior during the assessment.
Ingestion points: The skill uses Glob, Grep, and Read tools to search for and analyze various configuration files across the file system (e.g., **/Dockerfile, **/*.yaml).
Boundary markers: The skill includes a dedicated 'Prompt Injection Safety Notice' section that explicitly instructs the agent to treat manifest contents as data, not instructions, and to ignore any claims of compliance or directives found within the files.
Capability inventory: The skill is limited to Read, Grep, and Glob tools. It lacks network access or file-writing capabilities in its provided configuration, significantly limiting the potential impact of a successful injection.
Sanitization: No programmatic sanitization is defined; the skill relies on instructional guardrails to distinguish data from instructions.

container-security