PicoClaw — agentic threat model

Relies entirely on external LLM providers via API keys. Vulnerable to prompt injection attacks that could manipulate the bot's responses or trigger unintended automation actions, as well as potential API key abuse if keys are intercepted.

Not certain from the listing — No local vector database or RAG operations are mentioned; the agent likely relies entirely on the external LLM's context window or basic in-memory message history.

Not certain from the listing — The orchestration framework is a lightweight Go binary. While it supports personal automation, the extent of its tool-calling capabilities, memory management, and planning logic is not detailed.

Runs as a single portable Go binary on embedded Linux (including RISC-V). The primary threat is host compromise if the binary is run with root privileges, combined with the risk of cleartext API keys stored in the local configuration file.

Not certain from the listing — No built-in evaluation, guardrails, or observability tools are described; monitoring is likely limited to standard Go console logs or platform-specific gateway logs.

Lacks enterprise-grade security controls. Security relies entirely on the host operating system's file permissions to protect the configuration file containing LLM API keys and chat platform tokens.

Connects to external chat ecosystems (Telegram, Discord) via gateway components. Vulnerabilities include unauthorized users interacting with the bot if access control lists are not properly configured within the bot's settings.