The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external subtitle files (SRT and ASS) as part of its core functionality described in SKILL.md and references/workflow.md. This represents a potential surface for indirect prompt injection.
Ingestion points: Subtitle imports via jianying captions import-srt and import-ass commands.
Boundary markers: The skill uses structural validation for timing and overlaps, but does not explicitly instruct the agent to ignore or sanitize natural language instructions embedded within the subtitle text.
Capability inventory: The agent has access to the jianying CLI, which can modify, style, and export project data.
Sanitization: Validation is present for structural integrity (e.g., negative time, overlaps), but lacks specific sanitization for prompt-based content within the text payload.
[EXTERNAL_DOWNLOADS]: The workflow documentation in references/workflow.md suggests installing a complementary skill using the npx skills add full-aigc-skills/jianying-skills command. This resource is provided by the same author as the current skill and is intended for speech-to-text functionality.
[COMMAND_EXECUTION]: The skill operates by executing the jianying Rust CLI tool for subtitle management tasks, including listing, importing, editing, styling, and exporting, as detailed in references/workflow.md.

jianying-draft-subtitles