protocol-reverse-engineering

Fail

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: HIGHPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PERSISTENCE]: The skill defines a "Memory Protocol" that mandates the agent to read and write to specific absolute file paths (e.g., C:\dev\projects\agent-studio\.claude\context\memory\learnings.md). This mechanism is designed to persist "learnings," "issues," and "decisions" across different sessions, effectively overriding standard session boundaries and allowing the skill to maintain a stateful influence over the agent's behavior.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from network captures and live traffic.
  • Ingestion points: PCAP files (capture.pcap), live traffic via tshark, tcpdump, and mitmproxy (identified in SKILL.md).
  • Boundary markers: None present; the instructions do not provide delimiters or warnings to ignore instructions embedded within the network data.
  • Capability inventory: The agent has access to powerful tools including Bash for shell execution and Write/Edit for modifying local files (defined in SKILL.md frontmatter).
  • Sanitization: No sanitization or validation of the ingested network content is performed before processing.
  • [COMMAND_EXECUTION]: The instructions include usage of powerful command-line tools for network surveillance and traffic interception, such as tcpdump, tshark, and mitmproxy. While relevant to the skill's purpose, these tools provide significant capabilities for monitoring local and network activity.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 14, 2026, 09:49 PM