The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scanners (URLite and File Reputation) have identified the documentation URL ('https://jeffallan.github.io/claude-skills/skills/infrastructure/postgres-pro/') and the 'SKILL.md' file as malicious.
[COMMAND_EXECUTION]: The file 'references/replication.md' contains multiple instances of 'rm -rf /var/lib/postgresql/14/main/*', a highly destructive command that wipes the database data directory. While accompanied by text-based warnings, the inclusion of the raw command string poses a significant risk of catastrophic data loss if executed in a production environment.
[METADATA_POISONING]: There is a discrepancy between the stated author in the platform context ('emmraan') and the author metadata within 'SKILL.md' ('https://github.com/Jeffallan'). This suggests potential impersonation or deceptive attribution of the skill's source.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes dynamic data from database catalogs, system views, and query optimization tools (EXPLAIN ANALYZE) across several files including 'references/maintenance.md' and 'references/performance.md'.
Boundary markers: The instructions lack delimiters or explicit warnings to the agent to treat data retrieved from the database as untrusted content.
Capability inventory: The skill provides instructions for sensitive shell operations including service management ('systemctl'), file system modification ('rm', 'tar'), and specialized database tools ('pg_basebackup').
Sanitization: No validation or sanitization mechanisms are implemented for the data retrieved from the database before it is processed by the agent.
Remediation: Use structured data formats for query results, implement strict validation of database outputs, and replace destructive shell command strings with descriptive placeholders that require manual user confirmation and input.

postgres-pro