The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates using a local domain-specific CLI (jianying) to handle video project metadata. Execution is limited to supported capabilities and requires explicit user authorization for operations with potential side effects.\n- [DATA_EXFILTRATION]: No exfiltration risks were identified. Instructions explicitly prohibit the collection of extraneous credentials and restrict data processing to authorized local and provider-defined environments.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project drafts and media metadata (ingestion points in SKILL.md). It uses jianying-job/v2 schema boundaries and performs structural validation (sanitization) using CLI-based auditing capabilities (doctor, probe, verify). This surface is well-defended by mandatory check-and-verify workflows.\n- [SAFE]: The skill follows secure design principles, including mandatory isolation for file writes, a ban on generating executable scripts, and strict rules against automatic retries for ambiguous task states.

jianying-draft