The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external invention disclosures and patent drafts via the 'Read' tool. These documents constitute untrusted data that enters the agent's context. The skill lacks explicit instructions for boundary markers (e.g., delimiters) or sanitization protocols to prevent embedded instructions within these documents from hijacking the agent's flow.
[COMMAND_EXECUTION]: The skill leverages a 'Bash' tool to perform compliance checks. While standard for this use case, the combination of shell access with the processing of untrusted external documents increases the risk that a successful indirect prompt injection could lead to unauthorized command execution.

pct-application