The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted marketing content provided by the user to perform quality gates and voice pattern analysis.
Ingestion points: Content provided for scoring via triggers like "score this", "quality check", or "run quality gates".
Boundary markers: While the skill ignores HTML comments and code fences during regex matching, there are no explicit delimiters or "ignore embedded instructions" warnings for the primary content ingestion.
Capability inventory: The skill uses the grep tool for pattern matching and performs file-write operations to data/learning/gate_runs.jsonl and memory/lessons.md.
Sanitization: The instructions do not define sanitization, validation, or escaping procedures for the external content before it is processed or used to update the agent's memory (lessons). If a user crafts content to intentionally trigger repeated failures, they could influence the agent's persistent memory/lessons.md file.

kai-gate