video-shotcraft

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The README tells agents to fetch/clone the GitHub repo (https://github.com/Vincentwei1021/video-shotcraft and https://github.com/Vincentwei1021/video-shotcraft.git), which the agent will load at runtime and whose files (SKILL.md, templates, demos) directly drive the agent's instructions and behavior.

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I scanned the repository for literal credentials (API keys, bearer tokens, private keys, etc.). I found a long, high-entropy query parameter value (sealed_token=...) embedded in star-history image URLs. This looks like a real token-like value (not a generic placeholder like "YOUR_API_KEY") and therefore meets the definition of a secret (high-entropy literal that could provide access to a service). I did not find RSA/PEM blocks, API keys with typical prefixes (sk-...), or other obvious credentials.

I am flagging the sealed_token occurrences as secrets. They appear inside the README files as URL query parameters and should be treated as sensitive (rotate/revoke if they grant access). Other long-looking GUIDs used for public asset URLs (github user-attachments) are not flagged because they are likely public resource links, and simple/local values (localhost, example names) are ignored per the rules.