PopeBot: Secure Self-Improving AI Agent

You are an expert AI systems architect specializing in secure, self-improving autonomous agents. Design and implement a secure AI bot system with the following architecture: ## CORE ARCHITECTURE Build a self-hosted AI agent that operates through the following pipeline: 1. **Chat Interface** (Telegram bot + REST API webhook) for receiving job requests 2. **Job Orchestration** via GitHub Actions — each request spins up an isolated Docker container 3. **Configuration as Code** — all agent behaviors, skills, and instructions stored as Markdown files in an "operating system" folder within the Git repository 4. **Pull Request Workflow** — all agent-generated changes submitted as PRs for human review before merging 5. **Self-Improvement Loop** — agent logs stored in Git; future jobs analyze past logs to optimize performance ## SECURITY MODEL (Two-Tier Secret Management) Implement strict credential separation: ### Container Secrets (LLM CANNOT access): - GITHUB_TOKEN (repo operations) - ANTHROPIC_API_KEY (LLM inference) - WEBHOOK_SECRET (callback authentication) These are injected at runtime via GitHub Secrets → Docker environment variables. Never stored in code. ### LLM Secrets (LLM CAN access): - BRAVE_SEARCH_API_KEY (web research) - Any task-specific credentials the agent needs for its work Minimized to reduce attack surface. ### Security Principles: - No credentials in repository code or job definitions - Fresh isolated container per job (no shared state) - Webhook authentication for all callbacks - PR review gate before any changes merge to main branch ## JOB LIFECYCLE For each request: 1. User sends message via Telegram or API call 2. Bot creates an execution plan and presents it for approval 3. On approval → GitHub Action triggers → Docker container spins up 4. Secrets injected at runtime from GitHub Secrets 5. Agent executes task (has access to Brave Search, file creation, code modification) 6. All work submitted as a Pull Request with full logs 7. Completion report sent back to chat interface with PR links 8. User reviews and merges PR to apply changes ## SCHEDULED TASKS Support cron job definitions in a configuration file: - Each cron entry specifies: schedule (cron syntax), command, enabled/disabled status - Agent can create new cron jobs as part of task execution - Example: "Daily financial research at 6 AM → read instructions → generate report → commit to repo" ## SELF-IMPROVEMENT MECHANISM - All agent execution logs committed to Git - Create analysis jobs that review past logs for: - Failure patterns and error resolution - Performance bottlenecks - Successful strategies to replicate - Agent can modify its own operating system files to improve future performance ## OUTPUT REQUIREMENTS For any job, the agent should produce: 1. The job definition file (what was requested) 2. Complete execution logs (thought process, API calls, decisions) 3. The actual deliverable (report, code, configuration, etc.) 4. A summary message back to the chat interface Design this system to be forkable, configurable, and secure by default with optional auto-merge for trusted workflows.