rag-local-with-container
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill outlines a workflow for ingesting and processing text corpus data for embeddings and database storage (described in Step 2 of SKILL.md). Processing untrusted external data can introduce an indirect prompt injection surface where embedded instructions might influence downstream agent behavior. The skill mitigates this by utilizing parameterized SQL queries (
cursor.executewith?placeholders), which prevents direct command injection into the database layer. - [Local and External Network Interactions]: The skill documents interactions with a local embedding endpoint (Ollama) and includes a diagnostic probe to a well-known service (GitHub API) to demonstrate network allowlist differences between local and cloud environments. These interactions are transparently described and serve the skill's primary purpose of environment parity verification.
- [Credential Management]: The skill uses environment variables (
$SQL_PASSWORD) for database authentication in its testing examples, which is a standard and recommended security practice for avoiding hardcoded secrets.
Audit Metadata