The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, specifically slide content and presentation comments, via the Google Slides and Drive APIs. This content is provided to the agent without sanitization or explicit boundary markers, which could allow instructions embedded in user-generated presentation content to influence agent behavior (Indirect Prompt Injection).
Ingestion points: slides_skill.py fetches content using presentations().get() and comments().list().
Boundary markers: Absent in instructions.
Capability inventory: The skill allows arbitrary command execution via the Bash tool to run the Python script, and can write files (e.g., exported PDFs/PPTXs).
Sanitization: No filtering or sanitization of ingested content is performed.
[CREDENTIALS_UNSAFE]: The skill's setup script (slides_skill.py) is designed to search for and read credentials.json files in the directories of related skills (~/.claude/skills/gmail-skill/ and ~/.claude/skills/google-sheets-skill/). While this facilitates shared authentication state within the agent's environment, it involves accessing sensitive configuration files outside of the skill's own directory.
[COMMAND_EXECUTION]: The skill instructions require the agent to execute local Python scripts using the Bash tool to perform all presentation management tasks.
[SAFE]: The skill utilizes official Google API client libraries and communicates with Google's well-known and trusted infrastructure. The retrieval of slide thumbnails from Google-hosted URLs is a standard functional requirement for the skill's operation.

google-slides-skill