The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions direct the agent to read the CLAUDE.md file specifically to retrieve 'API keys'. While this file is a standard context file for the target platform, explicitly targeting credentials in a utility whose primary purpose (video prompt generation) does not require them is a security concern.
[PROMPT_INJECTION]: The skill implements a 'Self-Improving Loop' that records execution feedback into logs/last-run.md and reads it at the beginning of each session. This persistent data store creates a surface for indirect prompt injection. * Ingestion points: User-provided creative briefs and video descriptions (SKILL.md). * Boundary markers: Absent; there are no instructions to the agent to isolate user-provided text or ignore instructions embedded in the briefs. * Capability inventory: The skill uses file-write operations to update the log file and reads the file to modify agent behavior. * Sanitization: Absent; there is no validation or filtering of the feedback before it is persisted to the project history.
[EXTERNAL_DOWNLOADS]: The README provides installation instructions that fetch the skill's source and rule files from the vendor's official GitHub repository.

seedance-prompting