If you discover a security vulnerability within llmfetch, please send an email to the project maintainer. All reports will be promptly addressed.

llmfetch is designed with a minimal attack surface:

• No API keys or secrets: The project uses only local Ollama models. No external services are called, and no secrets are stored or transmitted.
• No user input handling: The pipeline only processes commit hashes (from GitHub Actions) and LLM output (from local Ollama). There is no web interface or user-facing input endpoint.
• No network calls: The only network call is to localhost:11434 (local Ollama). No external APIs are used.
• Deterministic output: The rendering pipeline is deterministic given the same commit hash, reducing the risk of injection attacks.

• The Ollama model output is parsed without strict validation; malformed output is handled gracefully but could theoretically contain unexpected characters.
• The pipeline runs with the same permissions as the GitHub Actions runner.