The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection due to its core functionality of processing untrusted external data.
Ingestion points: In SKILL.md, the instructions prompt the agent to analyze various external sources, including keyword exports, customer call transcripts, survey responses, and live web search results from Reddit, Quora, and competitor blogs.
Boundary markers: The skill lacks explicit delimiters or instructions to treat external data as untrusted, which could lead the agent to inadvertently follow instructions embedded within transcripts or web content.
Capability inventory: The skill does not possess dangerous capabilities; no subprocess calls, file-write operations (other than reading specific context files), or unauthorized network operations were detected.
Sanitization: There is no evidence of sanitization or filtering logic to prevent the interpretation of malicious payloads within the ingested content.
[NO_CODE]: The skill consists entirely of markdown instructions and JSON metadata. No executable code files, scripts, or binary assets were found within the provided files.

content-strategy