The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's primary purpose is to upload local files and folders to a remote service (https://www.agenticresearch.sh).
Evidence: scripts/drop.py reads file contents and sends them via urllib.request to the vendor's API.
Mitigations: The skill includes a looks_secret function and a SKIP_NAMES/SKIP_SUFFIXES list to prevent uploading sensitive files like .env, .ssh/id_rsa, .pem, and credentials. It also defaults to excluding hidden files.
[REMOTE_CODE_EXECUTION]: The documentation suggests a fallback method for users who haven't installed the skill, which involves piping a remote script directly into the Python interpreter.
Evidence: SKILL.md contains the instruction: `curl -fsSL https://www.agenticresearch.sh/s/drop.py | python3
`.
Analysis: While the target domain is the vendor's own site (agenticresearch.sh), piping remote content to a shell or interpreter is a high-risk pattern. However, since the source is the vendor itself and the skill provides a local copy of the script for standard use, this is a documented fallback rather than a primary malicious vector.
[COMMAND_EXECUTION]: The skill executes a local Python script to perform the upload, incorporating user-provided paths and titles.
Evidence: python3 "${CLAUDE_SKILL_DIR}/scripts/drop.py" <path> --title "<what this is>".
Analysis: This is standard skill operation. The script uses argparse and pathlib for safe path handling, minimizing command injection risks from user input.
[SAFE]: All network operations and remote references target the vendor's official domain (agenticresearch.sh), which is consistent with the skill's stated purpose.

context-drop