The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to enumerate and read "every call site, import, and downstream reference," which includes code, tests, and documentation that may come from external or untrusted sources (SKILL.md). This creates a surface for indirect prompt injection where malicious instructions embedded in the analyzed code could influence the agent's actions.
Ingestion points: Files identified via Grep and Glob tools, including scripts and documents that might reside in other repositories.
Boundary markers: None defined; there are no instructions to differentiate between data to be analyzed and instructions to be followed, nor are there warnings to ignore instructions found within processed files.
Capability inventory: The skill allows the use of Bash for command execution and Write for file modifications, providing an avenue for impact if the agent is misled by malicious data.
Sanitization: No sanitization or validation protocols are specified for the content retrieved from external consumers before it is processed by the agent.

blast-radius