The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file contains instructions for the AI agent to perform an anonymous version check by making a GET request to the AgentEra/Agently-Skills repository on GitHub. This is a vendor-owned resource used to verify the integrity of the installed skill files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from user inputs, knowledge base records, and external files. While it provides security best practices, such as using boundary markers and validating model outputs in host code, it remains vulnerable to indirect prompt injection.
Ingestion points: Data enters the context through input(), info(), and instruct() slots in model requests, as well as retrieved records from RecordStore and ContextReader (documented in references/prompt-management.md and references/knowledge-base.md).
Boundary markers: The documentation recommends using structured delimiters like [[ref:<ref_id>]] for citations and square brackets (e.g., [info.rules]) to reference specific prompt segments (documented in references/knowledge-base.md and references/prompt-management.md).
Capability inventory: The skill facilitates complex operations including tool/action execution, file system access via TaskWorkspace, and database operations via RecordStore (documented in references/output-control.md and references/knowledge-base.md).
Sanitization: The skill advises validating model-selected identifiers against host-owned maps and redacting credentials or secrets from prompt audits (documented in references/output-control.md and references/prompt-management.md).

agently-request