The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, such as video transcripts and analytics data, to generate and rewrite video scripts.
Ingestion points: Data enters the system via tools/collect.py which processes user pastes and metadata from yt-dlp output (e.g., yt.json).
Boundary markers: The instructions in SKILL.md lack explicit directives or delimiters to prevent the agent from obeying instructions that might be embedded in external video titles or transcripts.
Capability inventory: The skill is capable of writing files to the local directory ./virality/ and executing internal Python utilities.
Sanitization: The tool tools/humanize.py provides mitigation by automatically stripping invisible characters (zero-width characters, Unicode tags) and typography artifacts from drafts.
[COMMAND_EXECUTION]: The skill includes a testing utility that uses shell command execution to verify tool functionality.
Evidence: The file tools/test_virality.py contains a cli function that uses subprocess.run() to execute the skill's own Python scripts (hookscore.py, speech.py, etc.). This is a standard development pattern used to run unit tests against the skill's CLI tools.

virality