pp-public-param-golden

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements a self-learning loop (Category 8: Indirect Prompt Injection) that allows the agent to ingest and follow playbooks and notes stored in a local SQLite database.
  • Ingestion points: The agent reads choreography and guidance from the recall command results.
  • Boundary markers: SKILL.md provides explicit instructions to the agent to verify candidate actions and follow specific protocol steps before committing to a playbook.
  • Capability inventory: The skill can execute various CLI commands via the public-param-golden-pp-cli binary, including store creation and data lookups.
  • Sanitization: The instructions include a requirement for the agent to strip PII from queries and notes before teaching them to the system.
  • [DYNAMIC_EXECUTION]: The sql tool allows for the execution of arbitrary SQL queries against the local database. This is mitigated by a validateReadOnlyQuery function in internal/mcp/tools.go that restricts execution to SELECT and WITH statements to prevent data modification.
  • [EXTERNAL_DOWNLOADS]: The skill's documentation recommends installation via npx using the @mvanhorn/printing-press-library package. As this originates from the verified skill author's library, it is considered a legitimate installation path.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 07:39 AM
Security Audit — agent-trust-hub — pp-public-param-golden