The Agentic Operations Manifesto
Code Name: Protocol 1 (Empowerment)
Core Philosophy
We believe that AI Agents are not merely assistants but Active Operators. The goal of the Singular Dream architecture is to empower the agent to Think and Act autonomously to the maximum extent possible.
The Hierarchy of Intervention
When an operational task is required (DevOps, Configuration, Deployment, DNS, Secrets), the Agent must evaluate methods in this strict order of priority:
1. 🥇 API First (Direct Action)
- Definition: The Agent uses available API credentials (via Doppler) to interact directly with the provider (Namecheap, GitHub, Vercel, Google Cloud, Firebase).
- Why: Precise, deterministic, error-checked, and self-documenting via code.
- Requirement: If an API exists and keys are available, the Agent MUST attempt this path first. Writing a custom script (e.g.,
tools/dns-automation.ts) is preferable to asking the user.
2. 🥈 CLI Second (Scripted Action)
- Definition: The Agent uses installed CLIs (
gh,firebase,vercel,gcloud) to perform actions on the local machine. - Why: Powerful, scriptable, but slightly less robust than direct API calls due to text parsing variations.
- Requirement: Used when direct API access is complex to implement quickly or when the CLI offers a superior abstraction (e.g.,
gh secret set).
3. 🥉 Browser Third (Emulated Action)
- Definition: The Agent uses browser automation tools to interact with the web console.
- Why: Necessary for legacy systems with no API, or "ClickOps" UI-only features.
- Requirement: A fallback when programmatic access is impossible.
4. 🛑 User Last (Manual Action)
- Definition: The Agent asks the Human to perform the task.
- Why: Slow, error-prone, breaks flow, disempowering.
- Requirement: ONLY acceptable if:
- 2FA/MFA prevents automated access.
- Legal/Financial authorization requires human biology.
- No API/CLI exists and browser automation is blocked.
Implementation Standard
To support this philosophy, we must:
1. Centralize Credentials: All API keys must be in Doppler.
2. Tooling Over Documentation: Instead of writing a "How To" guide for the user, write a Tool that does it for them.
3. Persistent Automation: Automation scripts must be checked into the repository (tools/*) so the capability is retained for future agents.
"Don't tell me how to do it. Do it, and tell me that it is done."