eagle3-review-logs

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell utilities like find, ls, and tail to access and read experiment logs. The use of a variable in the path experiments/<exp_id>/ could potentially be exploited for directory traversal if the experiment ID is not validated.
  • [COMMAND_EXECUTION]: The skill suggests re-running tasks using uv run launch.py with the --yes flag, which instructs the agent to skip interactive confirmation prompts.
  • [PROMPT_INJECTION]: The skill reads and summarizes external log files, creating a surface for indirect prompt injection where malicious content in logs could attempt to manipulate the agent's summary or report.
  • Ingestion points: Log files matching *.out or *.log within experiment subdirectories found via the find command.
  • Boundary markers: None. The logs are read and processed directly without delimiters.
  • Capability inventory: File system access (ls, find), file reading (tail), and command execution (uv run).
  • Sanitization: None.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 01:46 PM
Security Audit — agent-trust-hub — eagle3-review-logs