Yggnagisanzenin/skyclaw
↗ GitHubRadically Innovative AI Agent. Free and Open Source Forever.
387
Stars
84
Forks
8
Watchers
6
Open Issues
Safety Rating A
The repository presents as a legitimate, well-documented open source AI agent project. Security architecture is notably thoughtful: secrets are stored in an encrypted vault with a 'vault://' URI scheme, API keys are auto-deleted from chat history, credentials are zeroized on drop, and access control is deny-by-default. No hardcoded secrets, obfuscated code, or malicious patterns are visible in the provided content. The capabilities described (computer use, shell execution, web browsing with credential handling) are powerful but are explicitly documented as user-controlled features. No prompt injection attempts targeting this analysis were detected.
ℹAI-assisted review, not a professional security audit.
AI Analysis
TEMM1E (Tem) is an autonomous AI agent runtime written in Rust, designed to run persistently on a server and interact via Telegram, Discord, WhatsApp, Slack, or a CLI/TUI. It features a multi-crate workspace architecture with a token-budget-aware agentic loop ('Tem's Mind'), exponential decay memory ('λ-Memory'), procedural blueprint system, swarm intelligence ('Many Tems'), full computer/desktop control ('Tem Gaze'), web browsing with OTK authentication ('Tem Prowl'), a self-tuning fine-tuning pipeline ('Eigen-Tune'), and a consciousness/observer layer. Supports 8 LLM providers including Anthropic, OpenAI, Gemini, xAI, and OpenRouter.
Use Cases
- Deploying a persistent autonomous AI agent on a VPS that responds to messages across Telegram, Discord, WhatsApp, and Slack
- Automating web browsing and desktop GUI interactions using vision-based computer control
- Running multi-agent swarm tasks in parallel with stigmergic coordination
- Performing long-horizon agentic tasks with procedural memory (blueprints) and cross-session recall
- Self-tuning a local LLM on captured interaction data with zero additional API cost
Tags
Security Findings (2)
No hardcoded secrets detected in the README or visible configuration. Secrets are handled via environment variables and an encrypted vault (ChaCha20-Poly1305). This appears intentional and well-designed.
No prompt injection attempts detected. The README contains standard technical documentation with no embedded instructions targeting AI analysts.
Project Connections
Model Context Protocol (MCP)
TEMM1E includes a built-in MCP client crate (temm1e-mcp) with a 14-server registry and supports discovering and installing MCP servers at runtime, making it a consumer of the MCP ecosystem.
Letta (formerly MemGPT)
Both projects implement persistent, memory-augmented AI agents with cross-session recall. TEMM1E explicitly benchmarks its λ-Memory system against Letta in its research documentation.
OpenTelemetry
TEMM1E includes a temm1e-observable crate that integrates OpenTelemetry for metrics and observability across the agent runtime.
Mem0
TEMM1E's λ-Memory system is positioned as an alternative to Mem0's memory management approach, with explicit competitive benchmarking included in the repository's research lab.
AutoGen / AG2
Both implement multi-agent orchestration frameworks. TEMM1E's 'Many Tems' swarm system offers an alternative coordination model (stigmergic/pheromone-based via SQLite) compared to AutoGen's LLM-to-LLM communication approach.