The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a significant surface for indirect prompt injection due to its reliance on raw, untrusted data processing.
Ingestion points: The skill specifically ingests data from external and potentially adversarial sources, including support ticket exports, NPS verbatims, app store reviews from third-party platforms, churn surveys, and social media mentions, as defined in Step 1 of SKILL.md.
Boundary markers: The instructions do not define delimiters (such as XML tags or triple backticks) or provide explicit warnings to the agent to ignore any natural language instructions embedded within the feedback items.
Capability inventory: The skill's allowed-tools list includes Bash, Write, Edit, and WebFetch, which represent a high-impact capability set if the agent were manipulated by an injection attack.
Sanitization: No validation or filtering mechanisms are described to identify or neutralize malicious payloads hidden within the feedback verbatims.

echo-feedback