show-system-prompt
Fail
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: HIGHPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFE
Full Analysis
- [PROMPT_INJECTION]: The core purpose of the skill is to "leak", "dump", and "extract" the runtime system prompt of AI agent CLIs. It explicitly targets vendor-confidential configuration data and provides multiple bypass methods to circumvent standard agent safety training.
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to run
npx --yes tweakcc, which downloads and executes a third-party script from the npm registry. This source is not verified or part of the trusted vendor list. - [COMMAND_EXECUTION]: The instructions require extensive use of powerful shell commands to inspect binary files, manipulate environment variables, and manage system-level certificates. This includes the use of
sudoto modify the system keychain. - [DATA_EXFILTRATION]: The skill configures the environment to capture raw API request bodies using
OTEL_LOG_RAW_API_BODIES. These logs, which typically contain sensitive API keys and session data, are written to world-readable directories like/tmp/claude-otel/. - [CREDENTIALS_UNSAFE]: By intercepting network traffic with
mitmproxyand dumping raw API bodies to disk, the skill creates a high risk of exposing active API credentials (e.g., Anthropic API keys) to local users or other processes.
Recommendations
- AI detected serious security threats
Audit Metadata