The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to target and extract highly sensitive data, including browser-stored passwords (Chrome Login Data), Windows Vault secrets, LSA secrets, and the Linux shadow file. These procedures are documented in SKILL.md and automated in scripts/agent.py.
[EXTERNAL_DOWNLOADS]: Instructions in SKILL.md and api-reference.md require downloading the LaZagne tool from an external GitHub repository (AlessandroZ/LaZagne). This source is not on the trusted vendor list, presenting a supply chain risk through the execution of third-party scripts and binaries.
[COMMAND_EXECUTION]: The skill uses shell commands for tool installation and execution. Additionally, scripts/agent.py uses the Python subprocess module to execute PowerShell commands for system auditing, specifically using Get-WinEvent to query security logs.
[PRIVILEGE_ESCALATION]: The skill provides instructions for escalating privileges to SYSTEM or root (via sudo) to access protected credential stores and master keys, which is a high-risk operation even in authorized testing contexts.
[INDIRECT_PROMPT_INJECTION]: The scripts/agent.py and scripts/process.py ingest untrusted system data (filenames, event log entries, and tool output) and process them into reports. These reports are presented to the agent, creating a surface where an attacker could influence agent behavior by manipulating local system artifacts.
Ingestion points: scripts/agent.py (filesystem metadata, Windows event logs), scripts/process.py (LaZagne JSON output files).
Boundary markers: None present in the processing scripts to distinguish between data and instructions.
Capability inventory: Subprocess execution (scripts/agent.py), file system writing (scripts/agent.py, scripts/process.py).
Sanitization: The scripts do not perform validation or sanitization of the ingested data before including it in the output summaries.

performing-credential-access-with-lazagne