The Agent Skills Directory

[NO_CODE]: This skill is composed entirely of Markdown instructions and routing metadata. It does not include any scripts (Python, Node.js, Shell), binaries, or direct tool execution commands. Its primary function is intent classification and routing.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied natural language to determine routing between sub-skills. While this represents an ingestion point for untrusted data, the skill possesses no capabilities (file writes, network calls, or command execution) to exploit. It lacks boundary markers for user input but is non-executable by design.
Ingestion points: User natural language requests processed by the routing algorithm in SKILL.md.
Boundary markers: Absent; the skill relies on keyword matching from the prompt.
Capability inventory: None; no subprocess calls, file operations, or network tools are defined in this skill.
Sanitization: None; input is used only for matching routing trigger phrases.
[DATA_EXPOSURE]: The skill mentions sensitive keywords such as huggingface_token, Secrets, and secure config store. These are used exclusively as trigger phrases to route the user to a setup sub-skill (exasol-ai-setup) and do not constitute hardcoded credentials or data exfiltration mechanisms.

exasol