The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it processes untrusted data from the web using WebSearch and WebFetch (detailed in references/company-research.md). Adversarial instructions on external sites could attempt to mislead the agent; however, the skill significantly reduces this risk through a mandatory 'Traceability Check' and 'Hallucination Red Flag Scan' defined in references/reflect-validate.md and references/verified-content-guardrails.md. These mechanisms ensure that only user-confirmed facts are included in the final output, effectively ignoring unverified external instructions.
[DATA_EXFILTRATION]: The skill accesses sensitive personal files such as master-facts.md and career-helper-preferences.md within the working directory. This access is limited to the skill's primary function of document personalization and accessibility adjustment. Analysis of the search and fetch patterns shows that personal history and metrics are kept local to the session and are not exfiltrated to external domains during the research phases.

application-optimiser