The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including user feedback, product reviews, support tickets, and social media comments, across all its core recipes (sentiment, classify, thematic, review).
Ingestion points: Untrusted data enters the agent context primarily through the review, sentiment, and classify subcommands as described in SKILL.md and reference/multi-channel-synthesis.md.
Boundary markers: The instructions do not specify the use of robust delimiters (e.g., triple-backticks, XML tags) or specific 'ignore instructions within data' warnings when processing verbatim feedback strings.
Capability inventory: The skill possesses the ability to 'recommend owners', 'write insight reports', and route findings to other agents like Growth, Spark, or Scout. While it does not execute shell commands, malicious instructions embedded in feedback could potentially influence the agent's summary or downstream recommendations.
Sanitization: No explicit sanitization or filtering of external content is mentioned before the data is interpolated into the analysis prompts.

voice