The Agent Skills Directory

Subprocess Execution for System Utilities: The skill uses Python's subprocess module to interact with system utilities such as git, ffprobe, and ffmpeg. These calls are implemented using argument lists rather than shell strings, which is a secure practice to prevent command injection.
Automated Artifact Retrieval: The artifact_lookup.py script retrieves artifacts from GitHub Actions and GitHub Pages. It enforces HTTPS and includes validation logic to ensure that only expected files are processed, mitigating risks from untrusted remote content.
Indirect Prompt Injection Mitigations: The skill contains explicit directives to treat researched source text and captured screens strictly as data. It instructs the agent to ignore any instructions or directives embedded within processed content.
Ingestion points: Untrusted data enters the context through files discovered by the rpi-research skill within user-defined source_roots.
Boundary markers: Instructions in SKILL.md and references/curriculum.md act as logical boundaries by directing the agent to ignore embedded directives.
Capability inventory: The skill uses subprocess calls for media processing and file writing to the level working directory, as seen in scripts/render_checks.py.
Sanitization: references/curriculum.md defines Evidence Rules for identifying and noting untrusted content within sources.
Secure Handling of Service Credentials: The skill provides specific guidance on the use of Azure AI Speech credentials, directing that authentication keys and resource IDs be kept out of manifests and execution logs.

hve-demo-material