The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content which creates a potential surface for indirect prompt injection attacks.
Ingestion points: The skill reads a draft file provided as an argument and source materials (files or URLs) provided via the --source flag in SKILL.md.
Boundary markers: The skill does not explicitly define sanitization or boundary markers for the ingested text, although it utilizes a 'fresh context' for the sub-agent to mitigate confirmation bias.
Capability inventory: The skill uses Read, Grep, and Glob for ingestion, Agent and Task to spawn a sub-agent, and Write to potentially regenerate sections of the draft.
Sanitization: There are no documented sanitization steps for the input draft or source material before they are passed to the agent context or the claim-verifier sub-agent.

verify-claims