The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection through its cultural reference retrieval system.
Ingestion points: The protocol protocols/copy-neta-injection.md and standard standards/neta-websearch-pipeline.md instruct the agent to use WebSearch to retrieve content from third-party platforms including X (Twitter), Reddit, TikTok, niconico, 2ch, and 5ch based on user-supplied topics.
Boundary markers: The instructions lack explicit delimiters or warnings to treat the retrieved content as untrusted data or to ignore instructions embedded within the retrieved snippets.
Capability inventory: The worker agent possesses WebSearch and the ability to dispatch multiple subagents as seen in protocols/copy-ideation-parallel.md.
Sanitization: No sanitization or filtering is performed on the retrieved text before it is passed to the reasoning stage (Phase B) for structural deconstruction.
[EXTERNAL_DOWNLOADS]: The skill references several external domains for information gathering and verification.
The standards/neta-websearch-pipeline.md standardizes the use of site: search operators on well-known social media services (Twitter, Reddit, LinkedIn) to gather cultural context.
The standards/neta-source-taxonomy.md standard provides an allow-list of archival and academic sources for literary verification, including aozora.gr.jp, gutenberg.org, and archive.org. These network operations are for data retrieval only and do not involve code execution.

copywriting-team