The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: Analysis of the vulnerability surface associated with processing external infrastructure files.
Ingestion points: The skill uses Glob and Read tools to ingest untrusted Infrastructure as Code configuration files (.tf, .yaml, .json, .bicep, etc.) as described in SKILL.md Step 1.
Boundary markers: Present. A robust 'Prompt Injection Safety Notice' is included in SKILL.md, explicitly directing the agent to treat all template content, comments, and tags strictly as data and to ignore any embedded directives or rule overrides.
Capability inventory: Extremely limited. The skill defines allowed-tools: Read, Grep, Glob in its frontmatter, ensuring it lacks the capability to write files, execute system commands, or perform network operations.
Sanitization: Lacks structural validation or automated sanitization of data contents, relying entirely on the provided LLM boundary instructions to mitigate parsing risks.

iac-security