The Constitution

Articles governing every aspect of the chati.dev orchestration system, from agent behavior and security to quality standards and memory management.

AI Agents Without Rules

Give an AI agent full autonomy and watch chaos unfold. Without explicit governance, agents make inconsistent decisions, ignore security, skip quality checks, and produce outputs that contradict each other. The more autonomous the system, the more you need clear rules.

How Chati.dev solves this

Enforcement Architecture

Articles flow through 4 enforcement levels into runtime hooks that validate every agent action in the pipeline.

Constitution enforcement architecture: articles to hooks to pipeline
BLOCK1

Non-negotiable: violations halt the pipeline

STRICT0

Enforced with safety net: requires remediation

GUIDE18

Recommended pattern: flexible implementation

WARN0

Advisory: attention required

Articles

Each article defines governance rules with specific enforcement levels. Click to expand.

Universal Protocols

Eight protocols that every agent follows, ensuring consistent behavior from discovery through deployment.

Agent execution protocol: self-validate, loop, handoff, gate
1

Self-Validation

Agents validate their own output against acceptance criteria before producing a handoff.

2

Loop Until Done

Agents iterate until all criteria are met. No partial handoffs are allowed.

3

Guided Options

Critical decisions are presented as numbered options (1, 2, 3) for user selection.

4

Two-Layer Handoff

Layer 1: structured YAML frontmatter. Layer 2: detailed artifact content with rationale.

5

Persistence

Agents retry on failure and investigate root cause before escalating. Max 3-5 retries depending on context.

6

Language Protocol

All artifacts in English. User interaction follows session.yaml language setting. First step output always in English.

7

Deviation Protocol

Scope changes, approach changes, skip/add agent requests are routed to the orchestrator. Requires user confirmation + audit trail.

8

Interaction Model

Agents lead the conversation proactively. They know their mission from the handoff and never ask 'what do you want me to do?'

Constitutional Framework

The Constitution is the highest authority in the system. All agents, hooks, and workflows must comply. The L0 layer in PRISM ensures these rules are always loaded, even in CRITICAL bracket.

19
Total Articles
1
BLOCK Level
18
GUIDE + WARN Level