The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external firewall configuration files, security group definitions, and infrastructure-as-code files which may contain attacker-controlled comments or metadata. It also interpolates user-provided $ARGUMENTS into the instructional text.
Ingestion points: Processes various local configuration files discovered via patterns like **/iptables*, **/*.tf, and **/*.yaml (defined in SKILL.md).
Boundary markers: The skill includes an explicit 'Prompt Injection Safety Notice' section that instructs the agent to treat all configuration content as untrusted data and ignore instructions found within comments or descriptions.
Capability inventory: Execution is scoped to read-only tools (Read, Grep, Glob) as defined in the allowed-tools frontmatter.
Sanitization: Relies on specific instructions to the agent to avoid evaluating or executing content found within analyzed files, rather than programmatic filtering.

firewall-review