skill-system-memory

Pass

Audited by Gen Agent Trust Hub on Mar 11, 2026

Risk Level: SAFECREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill manages database credentials by prompting the user and storing them in the standard PostgreSQL password file (.pgpass on Linux/macOS or pgpass.conf on Windows) using the scripts setup-pgpass.sh and setup-pgpass.ps1.\n- [COMMAND_EXECUTION]: Various scripts execute shell commands for system maintenance and setup. The bootstrap and setup-pgvector scripts use winget to install software and nmake to compile the pgvector extension from source code.\n- [EXTERNAL_DOWNLOADS]: The skill downloads the pgvector source code from the official GitHub repository and makes network requests to OpenAI or local Ollama endpoints to generate embeddings for stored memories.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface through its OpenCode plugin, which automatically retrieves stored content from the database and injects it into the agent's context during compaction operations.\n
  • Ingestion points: Data is stored in the agent_memories table via the store_memory function (init.sql) and retrieved in the plugin/skill-system-memory.js file.\n
  • Boundary markers: Retrieved memories are injected into the agent's context without the use of specific boundary markers or delimiters to isolate untrusted content.\n
  • Capability inventory: The skill is capable of performing database read/write operations (init.sql), executing shell commands via psql (mem.sh), and making network requests (ingest-embeddings.sh).\n
  • Sanitization: The skill implements SQL escaping to prevent database injection, but it does not sanitize the text content before it is injected into the prompt context.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 11, 2026, 12:52 PM