concurrency-exploitation

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documents numerous commands for conducting race condition attacks and system analysis, including process priority manipulation and CPU stressing.
  • Evidence: SKILL.md and payloads.md list tools like stress-ng, taskset, and strace for manipulating system behavior.
  • [DATA_EXPOSURE]: Documentation explicitly references sensitive system paths as targets for exploitation examples.
  • Evidence: SKILL.md and test-cases.md use /etc/shadow as a target example for symlink-based TOCTOU attacks.
  • [PRIVILEGE_ESCALATION]: The instructions include techniques for acquiring elevated privileges through process priority adjustment.
  • Evidence: guides/race-window-amplification.md suggests using sudo nice -n -20 to increase attacker process priority.
  • [INDIRECT_PROMPT_INJECTION]: The skill is configured to ingest external data from the web and system logs, creating a potential attack surface if malicious instructions are embedded in those sources.
  • Ingestion points: SKILL.md allows WebSearch and WebFetch tools; test-cases.md and guides/race-window-amplification.md instruct the agent to monitor audit logs and dmesg.
  • Boundary markers: Absent.
  • Capability inventory: The skill has Bash and Write permissions (SKILL.md).
  • Sanitization: Absent.
  • [DYNAMIC_EXECUTION]: The skill guides the user to compile and execute code at runtime to detect or trigger race conditions.
  • Evidence: SKILL.md and guides/threadsanitizer-guide.md provide patterns for compiling source code with gcc -fsanitize=thread and executing the resulting binary.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM