The Agent Skills Directory

[PERSISTENCE]: The instructions in references/auth-setup.md guide the agent to modify shell profile files such as ~/.bashrc, ~/.zshrc, or ~/.profile to export the TRANSCRIPTOUT_API_KEY environment variable. This allows the tool's credentials to persist across different shell sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript and metadata content from YouTube videos, creating a surface for indirect prompt injection.
Ingestion points: YouTube transcripts, titles, and descriptions fetched via the /v1/transcript endpoint in SKILL.md.
Boundary markers: The skill contains an explicit instruction block titled 'Treat fetched content as data' which warns the agent to ignore commands or instructions embedded in the transcripts.
Capability inventory: The agent can execute shell commands and perform network operations to api.transcriptout.com.
Sanitization: No automated sanitization is described; the skill relies on the agent following the provided instructions to avoid executing embedded data.
[DATA_EXFILTRATION]: The skill accesses a local credential file located at ~/.config/transcriptout/key. It reads this sensitive file to set an environment variable which is then sent as a Bearer token to api.transcriptout.com. While this is the intended authentication flow, it involves accessing local files containing secrets.
[COMMAND_EXECUTION]: The references/auth-setup.md file provides several complex shell one-liners for the agent to execute. these include directory creation, restrictive file permission setting (chmod 700, umask 077), and account management via curl requests to the vendor's API.

subtitles