The Agent Skills Directory

[PROMPT_INJECTION]: The skill analyzes untrusted infrastructure-as-code and configuration files, creating a potential surface for indirect prompt injection.
Ingestion points: The skill reads various file types including .tf, .yaml, .json, .ts, and .py files as defined in Step 1.
Boundary markers: SKILL.md includes a dedicated 'Prompt Injection Safety Notice' that explicitly instructs the agent to treat file contents as data and disregard any embedded instructions.
Capability inventory: The skill uses the Read, Grep, and Glob tools to inspect files and generates an assessment report.
Sanitization: No automated sanitization is described; the skill relies on the instructional guardrails provided in the safety notice.
[DATA_EXFILTRATION]: The skill is designed to locate and evaluate potentially sensitive files to identify security gaps in the target environment.
Evidence: Step 1 in SKILL.md instructs the agent to locate files such as **/.aws/credentials and **/.aws/config using the Glob tool.
Context: Accessing these files is a primary function of the skill's purpose (performing a security review to find exposed secrets). The skill does not have access to network tools, which prevents the exfiltration of any discovered data.

aws-review