The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill requires the agent to clone an external repository (github.com/guillaumemeyer/watermarks-remover) and execute its Python scripts directly. It also provides instructions for bootstrapping additional third-party repositories (aloshdenny/reverse-SynthID and mertizci/noai-watermark) to handle pixel-domain watermarks.
[EXTERNAL_DOWNLOADS]: The skill involves downloading external code and large binary models (~10 GB) from various third-party GitHub accounts. These sources are not associated with the skill author and are not recognized as well-known or trusted services.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from local files (PDF, DOCX, Images) and remote websites (audit_website.py). This data is then processed by a rewrite tool (rewrite_text.py) that uses an LLM, creating a surface for malicious instructions to influence the agent's behavior.
Ingestion points: The tool accepts local files via inspect_file.py and remote website content via audit_website.py.
Boundary markers: There are no mentioned delimiters or protective instructions used when passing external content to the LLM backend for rewriting.
Capability inventory: The skill has access to bash command execution, file system read/write operations, and network access for cloning repositories and auditing websites.
Sanitization: The analysis did not find evidence of input sanitization or filtering before external content is processed by the rewriting scripts.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute setup scripts, unified inspection tools, and unified cleaning tools that interact with the user's file system and network environment.

watermarks-remover