The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs runtime installation of several unversioned packages using pip3 install (git-dumper, ds_store) and npm install (source-map-explorer, unwebpack-sourcemap).\n- [REMOTE_CODE_EXECUTION]: The skill references and executes an external script (ds_store_exp.py) that must be obtained from a third-party GitHub repository (github.com/lijiejie/ds_store_exp), which is not a verified vendor.\n- [COMMAND_EXECUTION]: The skill makes extensive use of shell command execution, including piping output from network requests (curl) into language interpreters (python3) for data processing and extraction.\n- [DATA_EXFILTRATION]: The skill targets and downloads potentially sensitive files such as .env and .git repositories from remote targets into the local /tmp directory. While intended for reconnaissance, this pattern of fetching and storing sensitive data from external sources carries inherent risk.\n- [PROMPT_INJECTION]: The skill processes large volumes of untrusted data from external targets (HTML, JavaScript bundles, JSON manifests) using regular expressions and Python scripts. This creates an attack surface where malicious content from a target could potentially influence the agent's behavior.\n
Ingestion points: Data fetched via curl from remote targets (Phases 1, 2, 3, 5, 7, and 8).\n
Boundary markers: None identified; untrusted data is processed directly.\n
Capability inventory: Shell execution, filesystem access (/tmp), and network operations.\n
Sanitization: There is no evidence of sanitization or input validation for the data ingested from remote sources.

hunt-source-leak