The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process potentially untrusted business data, including AI architecture, customer-data sensitivity, and regulatory environments. This creates a surface for indirect prompt injection where adversarial instructions could be embedded in the business brief or risk evidence provided by a user. However, the skill's capabilities are restricted to read-only analysis and drafting within the agent's context, with no access to dangerous tools or network operations. The risk is mitigated by strict capability boundaries.
Ingestion points: Workflow steps in SKILL.md that require processing business model assumptions, contracts, and risk evidence.
Boundary markers: Absent; the skill relies on the agent's internal reasoning to handle input data.
Capability inventory: None; the skill does not invoke subprocesses, network calls, or file-write operations.
Sanitization: Not present; the skill treats all inputs as informational text for analysis.
[NO_CODE]: The skill consists entirely of Markdown-formatted instructions, templates, and reference links. It does not include any Python scripts, Node.js code, shell commands, or other executable artifacts, significantly reducing the potential attack surface.

saas-ai-risk-and-stress-test