The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (Algolia records, user events, and request IDs) and has the capability to modify local configuration or source code via the Write and Edit tools.
Ingestion points: External data enters the context through dataset files, request IDs, and Algolia index records mentioned in SKILL.md.
Boundary markers: The instructions do not define explicit delimiters (like XML tags) to separate processed data from instructions, nor do they include warnings to the agent to ignore instructions embedded in the data.
Capability inventory: The skill uses Write and Edit tools to apply minimization changes and WebFetch to retrieve documentation as defined in SKILL.md.
Sanitization: There are no explicit instructions for the agent to sanitize or escape data before processing it or interpolating it into prompts, although there are instructions to remove sensitive fields from the data itself.
[EXTERNAL_DOWNLOADS]: The skill uses WebFetch to access documentation from Algolia's official website (algolia.com). This is a well-known service and the usage is restricted to documentation retrieval, conforming to safe practices.
[CREDENTIALS_UNSAFE]: The skill demonstrates good security posture by explicitly instructing the user and agent to use restricted keys and to never include raw secrets or personal identifiers in logs or code.

algolia-data-handling