ZPrompt Agency

ZPrompt Agency builds ZPrompt XML prompts using local prompting guides and a safety-first workflow.

What this does

• Single Prompt Architect agent orchestrated by Agency Swarm.
• Reads local guides under ./mnt via read-only LocalShellTool and GuideIngestor.
• Merges user context with guide notes using PromptContextAggregator, ExampleInjector, and SafetyChecklist.
• Assembles complete ZPrompt XML covering title, role, backstory, goals, instructions, example, context, important, ethical, and tool-defenition.
• Validates every prompt with ZPromptValidator before returning it along with usage notes.

Setup

1. Create and activate a virtual environment.
2. pip install -r requirements.txt
3. Copy .env.template to .env and set OPENAI_API_KEY (with gpt-5.1 access) plus any other required keys.

Usage flow

1. Place or update guide files in ./mnt.
2. Start the agency and describe the task, domain, risks, and constraints.
3. The agent pulls relevant guides, builds context and safety checklists, emits ZPrompt XML, validates it, and returns the prompt with a short rationale.

Notes

• Tools are read-only against the repo and enforce path checks to keep operations inside ./mnt.
• Lists in the generated XML are trimmed and deduplicated; required sections must be non-empty for validation to pass.