claude-permissions-audit

Warn

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill reads and modifies multiple configuration files, including ~/.claude/settings.json and project-local settings, which directly control the agent's operational permissions and tool allowlists.
  • [CREDENTIALS_UNSAFE]: Accesses ~/.claude/settings.json, a security-sensitive file that defines tool permissions and may contain configuration for Model Context Protocol (MCP) servers, including authentication tokens or service-specific secrets.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) from untrusted configuration data.
  • Ingestion points: Reads settings from potentially untrusted project directories (e.g., /.claude/settings.json).
  • Boundary markers: Absent; the skill does not implement delimiters or safety instructions to isolate external configuration content from the agent's logic.
  • Capability inventory: Possesses file-writing capabilities to modify both global and local settings files, allowing it to apply changes to the agent's security posture.
  • Sanitization: Absent; the instructions do not include steps to validate or sanitize the content of ingested configuration files before processing and rewriting them.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 29, 2026, 04:33 PM