The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The helper script scripts/analyze_snapshot.py uses the pickle module to load memory snapshots provided by the user. The pickle.load() function is inherently unsafe as it can execute arbitrary code embedded within a malicious snapshot file. While the skill instructions advise only unpickling trusted files, the tool provided in the skill implement this unsafe deserialization, creating a high-risk vector for code execution if a user or an attacker provides a compromised .pickle or .pkl file.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted snapshot files.
Ingestion points: scripts/analyze_snapshot.py ingests data from external .json and .pickle snapshot files.
Boundary markers: The skill does not provide boundary markers or instructions to the agent to ignore or isolate instructions that may be embedded within the snapshot data (such as stack frame names or file paths).
Capability inventory: The agent has the ability to execute terminal commands, manage files, and potentially perform network operations.
Sanitization: The output from the analyzer script is not sanitized, meaning any malicious instructions within the snapshot will be presented directly to the agent's context.
[COMMAND_EXECUTION]: The skill workflow involves the agent constructing and executing shell commands to run the scripts/analyze_snapshot.py script. This execution pattern, particularly when handling external file paths, increases the attack surface and provides a execution path for the RCE vulnerability found in the script's deserialization logic.

analyze-memory-snapshot