creative-memory
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple ingestion points including project configuration files, domain guides, and memory tiers. This creates a surface for indirect prompt injection where malicious instructions hidden in these files could influence the agent's behavior during summary generation or knowledge updates.
- Ingestion points: Reads configuration from '.creative/config.yaml', domain-specific data from '$DOMAIN_DIR/domain.yaml', and instructions from '$DOMAIN_DIR/memory-guide.md'. It also reads recent content units for summarization (SKILL.md).
- Boundary markers: The skill does not define explicit boundary markers or delimiters for the external content it interpolates into prompts (SKILL.md).
- Capability inventory: Performs file read/write operations across all tiers and executes shell commands via grep for search functionality (SKILL.md).
- Sanitization: Includes checks for YAML frontmatter validity and syntax but lacks sanitization for the natural language content processed from memory files (SKILL.md).
- [COMMAND_EXECUTION]: The skill invokes the 'grep' utility to perform keyword searches across memory directories.
- Evidence: 'Use Grep to search memory directories (paths from domain config)' in the Query Memory execution flow (SKILL.md).
- [DYNAMIC_EXECUTION]: The skill dynamically computes file system paths (PLUGIN_ROOT and DOMAIN_DIR) at runtime based on its installation location and values retrieved from external configuration files.
- Evidence: Initialization sequence describing the calculation of 'PLUGIN_ROOT' and 'DOMAIN_DIR' (SKILL.md).
Audit Metadata