Safe Execution Framework for Autonomous Coding Agents

Stop AI agents from eating your infrastructure.

Note: This repository is currently a placeholder while the core SEFACA code is being developed in an incubator. The installation and usage instructions below represent the planned functionality.

curl -sSL https://sefaca.dev/install.sh | sh

sefaca run --context "[builder:ai:$USER@local(myapp:main)]" "python ai_agent.py"

• Execution Context: [persona:agent:reviewer@env(repo:branch)] - Complete tracking of every AI action
• Resource Limits: CPU, memory, and process constraints
• Pattern Detection: Blocks dangerous operations before execution
• Audit Trail: Every action logged and traceable

SEFACA uses a structured context format to track and identify every AI action:

[persona:agent:reviewer@environment(repository:branch)]

• persona: The role or type of agent (e.g., builder, reviewer, tester)
• agent: The AI system identifier (e.g., ai, gpt4, claude)
• reviewer: The human or system reviewing the actions (e.g., $USER, ci-bot)
• environment: The execution environment (e.g., local, staging, prod)
• repository: The code repository being worked on
• branch: The git branch being modified

Example: [builder:ai:jwalsh@local(myapp:feature-123)]

Documentation is being developed. See docs/README.md for planned content.

MIT

For more information, visit sefaca.dev