The Agent Skills Directory

[SAFE]: The skill defines a declarative workflow for podcast automation using YAML-based configurations and Handlebars-style templates (e.g., {{episode_title}}, {{audio_url}}). These are standard patterns for workflow automation.
[INDIRECT_PROMPT_INJECTION]: The skill defines several ingestion surfaces for untrusted data, such as episode summaries, guest biographies, and show notes templates. While these fields are interpolated into prompts during the publishing and distribution process, the risk is mitigated by standard model safety guardrails and the absence of high-privilege execution tools that could be abused via text injection. No specific evidence of exploitation surface beyond standard templating was found.
[COMMAND_EXECUTION]: All referenced tools (e.g., audio_process, podcast_publish) are managed through a specific MCP server (media-mcp) and do not involve direct shell access or arbitrary command execution within the skill definition.
[DATA_EXFILTRATION]: The skill describes distribution to well-known platforms (Apple Podcasts, Spotify, YouTube). These network operations are intrinsic to the stated purpose of the skill and utilize expected API endpoints, representing legitimate functionality rather than unauthorized exfiltration.

Podcast Automation