The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data (query, context, and pattern) from a calling skill to construct instructions for the Hex Threads Agent. This creates a surface for indirect prompt injection if the calling skill fails to sanitize user-provided input before passing it to this primitive.
Ingestion points: The query, context, and pattern inputs defined in the SKILL.md frontmatter and instructions.
Boundary markers: The skill does not employ explicit delimiters or boundary markers (e.g., XML tags or specific tokens) to isolate the interpolated data within the constructed prompt.
Capability inventory: The skill has the capability to create threads, poll for results, and extract data via the create_thread, get_thread, and continue_thread tools.
Sanitization: A specific guardrail is included in the documentation ("Do not pass raw user input directly as the Hex prompt"), but the skill relies on the calling skill to implement this sanitization correctly.
[DYNAMIC_EXECUTION]: The skill is designed to facilitate the execution of raw SQL queries provided by calling components at runtime.
Evidence: The "Rules" section under "Prompt Construction" explicitly directs the agent to "Pass raw SQL when available" and to "pass it directly to create_thread."
Mitigation: The skill includes a "Query safety" guardrail that instructs the agent to refuse any operations implying mutation (e.g., INSERT, UPDATE, DELETE), effectively enforcing a read-only constraint.

hex