The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection or safety bypass attempts was found. The instructions focus on maintaining accuracy and adhering to provided reference materials.
[DATA_EXFILTRATION]: The skill accesses a local configuration file (shinkoku.config.yaml) to identify database and output paths, which is standard for its tax-related functions. No network operations or attempts to exfiltrate sensitive data to external domains were identified.
[REMOTE_CODE_EXECUTION]: The skill does not perform any remote code execution or download external scripts. It operates entirely using local reference files and designated agent tools.
[OBFUSCATION]: A thorough scan for Base64, zero-width characters, homoglyphs, and other encoding techniques revealed no hidden or malicious content. External links, such as to the National Tax Agency (nta.go.jp), are plain-text and target official government resources.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user queries (untrusted input), it includes explicit instructions to use provided reference files as the sole source of truth and mandates a standard disclaimer for all responses, which serves as a functional boundary against manipulative inputs. Severity is evaluated as safe for this intended use case.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not utilize the dynamic context injection syntax (!command).

tax-advisor