The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from TikTok videos (such as titles, descriptions, or user comments) via the EchoTik API. This data is returned to the agent and could potentially contain malicious instructions.
Ingestion points: API responses in scripts/echotik_get_video_download_url.py are returned to the agent context.
Boundary markers: Prompt instructions in SKILL.md lack explicit delimiters or "ignore embedded instructions" warnings for the processed media metadata.
Capability inventory: The skill possesses file-write capabilities and the ability to execute subprocesses via scripts/response_io.py.
Sanitization: The script performs structural JSON validation but does not sanitize the textual content of the video metadata for injection patterns.
[PERSISTENCE]: The references/onboarding.md documentation contains instructions for users to modify their shell profiles (~/.zshrc, ~/.bashrc) to store the LINKFOX_AGENT_API_KEY environment variable. While these are provided as user instructions for environment setup, they define a persistence mechanism that an agent could potentially automate.
[COMMAND_EXECUTION]: The helper script scripts/response_io.py uses subprocess.run to execute other scripts within the skill package. While this is an architectural tool for managing large responses, it establishes a capability for command execution that relies on the parameters passed to the script.

linkfox-echotik-get-video-download-url