agent-guardrails-update

Warn

Audited by Gen Agent Trust Hub on Apr 15, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill dynamically generates a Python script at /tmp/agent-guardrails-update.py and executes it to process session logs.
  • [COMMAND_EXECUTION]: The skill modifies the .claude/hooks/stop-guardrails.sh shell script hook. Modifying such scripts allows for the persistent alteration of the agent's behavior and execution environment.
  • [DATA_EXFILTRATION]: The skill accesses session logs located in ~/.claude/projects/. These files contain historical interaction data and may include sensitive user information or code.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted session log data. \n
  • Ingestion points: Historical session logs in JSONL format from ~/.claude/projects/. \n
  • Boundary markers: The analysis process lacks explicit delimiters or instructions to ignore embedded commands within the log data. \n
  • Capability inventory: The skill can execute shell commands, write files, and edit local scripts. \n
  • Sanitization: No evidence of sanitization or validation of the log content is provided before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 15, 2026, 05:18 AM