The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external classroom transcripts and PPT data to generate notes. This presents an indirect prompt injection surface where malicious instructions embedded in a transcript could potentially influence the agent's behavior.
Ingestion points: The script scripts/maintain_obsidian_course.py reads transcript.txt, metadata.json, and ppt_outline.json from the replay output directory.
Boundary markers: There are no formal delimiters or specific safety warnings embedded within the ingested data files to prevent the agent from following instructions found in transcripts.
Capability inventory: The skill possesses the ability to execute shell commands via subprocess.run and write files to the local filesystem using Path.write_text.
Sanitization: Filenames are sanitized to remove illegal characters, but the text content of transcripts is used directly for summarization and note generation without content-level sanitization.
[COMMAND_EXECUTION]: The scripts/maintain_obsidian_course.py script executes external Python scripts (collect_buaa_course_replays.py and extract_buaa_classroom.py) located in a sibling project directory. This is intended for syncing course data but involves executing code from outside the skill's own package.

obsidian-course-vault