skill-system-insight

Fail

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The file scripts/evolve-soul.md contains a hardcoded PostgreSQL password (36795379) within a psql connection string example.
  • [COMMAND_EXECUTION]: The Python scripts scripts/cockpit_proposals.py and scripts/evolution_proposal.py utilize dynamic module loading via importlib.util from computed filesystem paths. The skill also performs database operations using the psycopg2 library and references the psql command-line utility.
  • [PROMPT_INJECTION]: The skill is designed to ingest untrusted session transcripts to evolve agent behavioral profiles and workflow recipes, creating a surface for indirect prompt injection. Ingestion points: Session transcripts processed via scripts/extract-facets.md. Boundary markers: None identified in the facet extraction or soul synthesis prompts. Capability inventory: File-write access to skill-system-soul/profiles/ and skill-system-workflow/recipes/, and database write access to memory tables. Sanitization: No evidence of content escaping or validation of user-controlled data before it is interpolated into agent instructions.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 16, 2026, 02:21 PM
Security Audit — agent-trust-hub — skill-system-insight