Skills
Navigate
skills/hiehoo/forex-rebate-bot/context-engineering/Gen Agent Trust Hub

context-engineering

Fail

Audited by Gen Agent Trust Hub on Apr 24, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The runtime awareness component documentation specifies procedures to read sensitive application credential files, such as ~/.claude/.credentials.json or the macOS Keychain, to harvest OAuth bearer tokens for API monitoring.
  • [COMMAND_EXECUTION]: The script scripts/tests/test_edge_cases.py utilizes subprocess.run to execute Python files on the local system during error handling verification.
  • [DATA_EXFILTRATION]: The skill facilitates the retrieval of account usage metrics from Anthropic's official API to provide real-time session awareness.
  • [PROMPT_INJECTION]: The scripts/context_analyzer.py and scripts/compression_evaluator.py scripts ingest untrusted message data (Ingestion: load_json_file) without boundary markers or sanitization, creating a surface for indirect prompt injection given the skill's command execution capabilities (Capability: subprocess.run in test_edge_cases.py).
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 24, 2026, 10:22 AM
Security Audit — agent-trust-hub — context-engineering