The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted configuration and template files to update its own instructions (SKILL.md).
Ingestion points: Reads versioned logic from files located in references/conductor-X.Y.Z/.
Boundary markers: Absent; there are no instructions to ignore or delimit embedded commands within the reference files.
Capability inventory: Extensive file-write capabilities across SKILL.md, the references/ directory, and the assets/ directory, combined with Git version control operations.
Sanitization: Absent; the skill lacks validation or sanitization logic for the content read from external references before updating the agent's operational logic.

conductor-upgrade