The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/agent.py and scripts/process.py execute external Windows forensic binaries (MFTECmd.exe, PECmd.exe, LECmd.exe, JLECmd.exe, SBECmd.exe, EvtxECmd.exe, RECmd.exe, RBCmd.exe, AmcacheParser.exe, AppCompatCacheParser.exe) using subprocess.run(). The arguments are handled safely as a list structure, which prevents shell command injection, but running local executables represents a high-privilege capability that relies on the integrity of the binaries folder path.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and aggregates untrusted forensic data from a target machine, which could contain attacker-controlled inputs designed to influence an AI agent analyzing the outputs.
Ingestion points: External files such as the Master File Table ($MFT), Windows Event Logs (.evtx), Prefetch files (.pf), LNK files, and Jump Lists are ingested via scripts/agent.py and scripts/process.py.
Boundary markers: No specific delimiters or boundary rules are implemented to isolate parsed artifact content from the broader context when ingested by an agent.
Capability inventory: The skill possesses powerful local capabilities including file-write operations for report generation and execution of external tools via subprocess.run().
Sanitization: There is no input sanitization or filtering of special characters applied to strings extracted from the forensic artifacts (such as FileName or event message strings) before they are written into JSON or CSV report timelines.

performing-windows-artifact-analysis-with-eric-zimmerman-tools