smart-init

Warn

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive local conversation history files stored in directories like ~/.claude/projects/ to extract patterns and corrections, exposing private user interaction logs to the agent's context.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted data from project files and conversation history to seed the 'Oracle' knowledge base.\n
  • Ingestion points: scripts/discover.py reads project files and interaction logs; SKILL.md instructs codebase and documentation analysis.\n
  • Boundary markers: No specific delimiters or warnings are used to prevent the agent from obeying instructions embedded within the analyzed data.\n
  • Capability inventory: The skill uses Write to create persistent knowledge entries and Bash to execute discovery scripts.\n
  • Sanitization: No sanitization or filtering of external content is performed before it is used to define project patterns.\n- [EXTERNAL_DOWNLOADS]: The skill recommends the installation of the @upstash/context7-mcp package from a well-known service to provide current library documentation.\n- [COMMAND_EXECUTION]: Executes a local Python script and various shell commands (find, grep, git) to gather information about the project environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 1, 2026, 05:29 PM