The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted artifacts, such as malware samples, firmware images, and suspicious documents, which inherently creates an indirect prompt injection surface.
Ingestion points: Multiple utility scripts (triage.py, string_scanner.py, entropy_scan.py, iat_analyzer.py) take untrusted file paths as arguments for processing.
Boundary markers: The skill documentation (e.g., SKILL.md, references/dynamic-analysis.md) strongly emphasizes the use of isolated lab environments, virtual machine snapshots, and network simulation to mitigate risks associated with untrusted data.
Capability inventory: The skill possesses the capability to read and write files, execute shell commands, and interact with the network for tool installation.
Sanitization: While the scripts parse raw binary data, the overarching methodology requires human-in-the-loop validation and execution within sandboxed environments.
[COMMAND_EXECUTION]: The setup_env.py script executes shell commands to facilitate the installation of necessary analysis tools.
Evidence: The script uses subprocess.run with shell=True to call system package managers such as apt-get on Linux and winget on Windows.
Purpose: This behavior is strictly for environment preparation, ensuring that standard analysis tools like gdb, yara, and binwalk are available to the agent.
[EXTERNAL_DOWNLOADS]: The environment setup process involves downloading tools and libraries from external sources.
Evidence: setup_env.py and references/install-guide.md reference the retrieval of packages from official system repositories and well-known registries like PyPI and NPM.
Context: These downloads target established service providers and are essential for the skill's functionality.

reversing-technique