proof-of-work

Warn

Audited by Gen Agent Trust Hub on Mar 11, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Recommends downloading and installing packages from the npm registry using npm install -g cclsp.- [REMOTE_CODE_EXECUTION]: Instructs the agent to run code directly from the npm registry using the npx cclsp@latest command.- [COMMAND_EXECUTION]: Utilizes a large suite of shell utilities and development tools (ps, cat, grep, jq, pytest, ruff, terraform, ansible) to validate environment state and implementation correctness. It also provides templates for creating executable Git pre-commit hooks.- [DATA_EXFILTRATION]: Employs curl to perform network operations, such as health checks on local services and testing API endpoints.- [PROMPT_INJECTION]: The skill processes untrusted output from external tools and shell commands, creating a surface for indirect prompt injection.
  • Ingestion points: Command output and logs captured as evidence in modules/evidence-logging.md.
  • Boundary markers: No delimiters are specified to isolate tool output from agent instructions.
  • Capability inventory: Extensive use of shell command execution (subprocess) throughout the skill modules.
  • Sanitization: No sanitization or validation of external tool output is defined.- [SAFE]: References official GitHub issue threads from the anthropics organization to provide context on known software bugs.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 11, 2026, 10:54 PM