postgres-pro

Fail

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references a documentation URL (https://jeffallan.github.io/claude-skills/skills/infrastructure/postgres-pro/) that has been blacklisted by automated security scanners (URLite).
  • [COMMAND_EXECUTION]: The core workflow and reference files contain powerful shell commands for database administration, including service management (systemctl), directory deletion (rm -rf), and binary backup tools (pg_basebackup).
  • [CREDENTIALS_UNSAFE]: Reference files references/replication.md and references/extensions.md provide configuration examples with hardcoded placeholder credentials (e.g., remote_password, secure_password, repl_password).
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from database queries and performance views, which constitutes an attack surface if the database content is sourced from untrusted parties. * Ingestion points: SKILL.md (EXPLAIN analysis), references/performance.md (query output analysis). * Boundary markers: Missing specific delimiters for untrusted database output. * Capability inventory: SQL execution, system shell access, file system modification. * Sanitization: Instructions recommend prepared statements, but do not provide specific sanitization for analyzing existing data logs.
  • [METADATA_POISONING]: There is a discrepancy between the system-recognized author ('emmraan') and the author metadata provided within the skill ('Jeffallan').
Recommendations
  • Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Sep 2, 2026, 01:00 AM
Security Audit — agent-trust-hub — postgres-pro