The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides code templates that process untrusted search queries, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The trackGleanSearch function within SKILL.md accepts an external query string.
Boundary markers: The instrumentation snippets lack explicit delimiters or instructions to ignore embedded commands within the processed query input.
Capability inventory: The skill frontmatter configuration allows access to Write, Edit, and Bash tools, which represents a significant capability tier if an injection were to occur.
Sanitization: The provided logging implementation includes a 100-character truncation for query strings, which acts as a rudimentary filter for large payloads.

glean-observability