The Agent Skills Directory

[COMMAND_EXECUTION]: The Python scripts scripts/go_binary_triage.py and scripts/macos_app_triage.py utilize subprocess.run to execute standard system utilities such as file, strings, otool, nm, and hdiutil. These tools are invoked with list-based arguments to prevent shell injection and are used exclusively for static metadata extraction and container triage.
[DATA_EXFILTRATION]: The skill handles potentially sensitive information during binary analysis. To prevent accidental exposure, it includes instructions to redact credential-like values and record only their existence and location. Triage artifacts are stored in a dedicated workspace under /tmp.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface when processing untrusted binaries. 1. Ingestion points: Data enters the agent context via triage scripts reading binary strings, symbols, and metadata. 2. Boundary markers: Instructions mandate a 'Default Posture' of static-first inspection, no-execution, and use of isolated workspaces. 3. Capability inventory: Scripts execute various CLI analysis tools via subprocess.run. 4. Sanitization: Arguments are passed as lists to prevent shell injection, and agents are advised to query artifacts selectively using tools like rg rather than displaying full contents in the conversation.
[SAFE]: The skill references established, well-known reverse-engineering tools and repositories such as mandiant/GoReSym and goretk/redress. The use of these external resources is consistent with the skill's legitimate purpose and does not involve suspicious download-and-pipe execution patterns.

reverse-engineering