The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated scans detected patterns where data from remote target endpoints is piped directly into a Python interpreter. In Phase 7, the skill retrieves content from the WooCommerce and Gravity Forms REST APIs and pipes the output to python3 for processing. While the skill uses the -c flag to provide a specific parsing script, this pattern introduces a risk if the remote data is maliciously crafted to exploit the script or interpreter environment.
[COMMAND_EXECUTION]: The skill relies extensively on high-privilege shell commands to perform its functions, utilizing tools like nmap, nuclei, sqlmap, httpx, and curl. It also explicitly instructs the user to perform privilege escalation using sudo systemctl start tor to facilitate anonymous scanning, which grants administrative control over the local system service.
[CREDENTIALS_UNSAFE]: The skill automates the discovery and extraction of hardcoded secrets from target infrastructure. Phase 2 extracts database credentials and API keys from server logs, while Phase 4 uses regex patterns to harvest AWS keys, Stripe live keys, and JWT tokens from JavaScript bundles.
[DATA_EXFILTRATION]: The skill performs automated data harvesting from target systems, collecting sensitive metadata such as server file paths, SQL query structures, and email addresses. This data is extracted and stored in temporary files on the local system (e.g., /tmp/error_log_$TARGET.txt).
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting and processing large volumes of untrusted data from external targets (logs, JS files, and API responses). This data is passed into high-capability tools without sanitization or boundary markers.
Ingestion points: Data is pulled from target log files (error_log, debug.log), JavaScript bundles, and various REST API endpoints.
Boundary markers: None; the skill does not include instructions for the agent to ignore or delimit instructions found within the ingested content.
Capability inventory: The agent has access to powerful network and exploitation tools like nmap and sqlmap which process this untrusted input.
Sanitization: None; ingested content is processed directly through regex or passed to other tools and interpreters.

deep-invade