The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted draft content which represents an indirect prompt injection surface. Maliciously crafted assertions in a draft could theoretically attempt to influence the subagent or the final reconciliation phase. However, the skill explicitly mitigates this by forking the subagent context and excluding the original draft text from its view.
Ingestion points: Draft file path provided in arguments.
Boundary markers: No explicit delimiters are mentioned for the extracted claims passed to the subagent.
Capability inventory: Uses Read for ingestion, Task for subagent orchestration, and Write for potential file regeneration.
Sanitization: No specific sanitization of extracted claim text is described.

verify-claims