The Agent Skills Directory

[PERSISTENCE]: The skill instructs the agent to store retrieved API keys in shell profiles such as ~/.bashrc or ~/.zshrc. This modifies the user's environment to maintain the credential across different sessions.
[DATA_EXFILTRATION]: Instructions explicitly guide the agent to circumvent platform-level security redaction. It suggests writing API responses to temporary files to extract keys if the platform's standard output filtering would otherwise hide the sensitive tokens from logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted YouTube transcript data via an external API, creating a surface for potential prompt injection.
Ingestion points: Video transcripts and metadata are retrieved from getyoutubetranscript.com and processed by the agent.
Boundary markers: The skill does not provide delimiters or specific instructions to the agent to treat the fetched content as untrusted or to ignore instructions embedded within the transcripts.
Capability inventory: The agent has the capability to execute shell commands (curl) and perform file system writes (modifying shell profiles and creating temp files).
Sanitization: No sanitization, escaping, or validation of the transcript content is performed before it is provided to the agent for analysis.
[COMMAND_EXECUTION]: The skill uses curl to perform various API operations, including a signup process that involves passing user-provided email addresses and verification codes directly into shell commands.

youtube-transcript