constitution

/constitution - Project Constitution

The Constitution owns durable governing principles, agent autonomy rules, Definition of Done, human checkpoints, and non-negotiables. It does not replace the Product Brief, PRD, MBB, specs, contracts, states, testing docs, workflow policies, or tier policy.

<input_contract> Read relevant evidence from:

• .memory-bank/constitution.md;
• .memory-bank/analysis/product-brief.md and the latest relevant brainstorm;
• .memory-bank/mbb/index.md, .memory-bank/spec-backbone.md, and .memory-bank/spec-index.md;
• .memory-bank/invariants.md and applicable contracts/states/testing/workflow docs;
• AGENTS.md and $ARGUMENTS.

If the Constitution is missing, create it from the current framework skeleton. Its minimum frontmatter remains:

description: Project Constitution - governing principles for AI-first development.
status: active
version: 1
project_principles: framework-default
ratified: null
last_updated: YYYY-MM-DD

Missing or blocked Product Brief does not prevent a generic governing-principles discussion, but it must be reported as context quality. </input_contract>

<hard_invariants>

• Ground project-specific principles only in accepted operator answers, explicit instructions, AGENTS.md, or existing authoritative project docs.
• Do not invent domain principles, Spec Kit hooks, command aliases, migration machinery, governance engines, or a second tier/status model.
• Do not silently weaken or override existing authoritative rules.
• An explicit operator skip is supported: keep framework-default or set skipped, record the skip, and allow /write-prd to continue. Absence of an answer is not an explicit skip. </hard_invariants>

<operator_decisions> Use an adaptive interview when a governing branch remains unresolved. Cover only the topics that lack authoritative answers, while ensuring the resulting Constitution addresses the applicable criteria:

• project level (enterprise|medium|junior); for non-enterprise work, an explicit DO NOT Overengineering-equivalent principle;
• architecture priority such as KISS or stability;
• Definition of Done and required checks;
• agent autonomy and human checkpoints;
• applicable critical non-negotiables such as security/privacy, no data loss, backward compatibility, performance, accessibility/UX consistency, low maintenance, or an operator-defined rule.

In interactive mode, ask exactly one most relevant unresolved governance question at a time. Offer exactly two mutually exclusive answer options plus свой вариант. Put the best-fit option first and end its text with the exact phrase Рекомендую этот вариант. Briefly explain the impact of each option.

After every accepted answer, apply it to the current Constitution context, reassess the remaining unresolved governance branches, and select the next most relevant question. Do not reveal a queued questionnaire. Do not impose a pass count or an artificial word limit on a custom answer.

Recommendations and suggested answers become authoritative only after explicit operator acceptance. In unattended use, apply only decisions already fixed by authoritative policy; otherwise stop with the active workflow's existing HALT_BLOCKING_QUESTIONS or HALT_CLARIFICATION_REQUIRED state. </operator_decisions>

<required_outputs> Ensure the Constitution reflects the current durable governing state from accepted evidence and sets project_principles:

• ratified: all applicable governing criteria have accepted answers;
• partial: some criteria remain open and the operator explicitly chooses to proceed;
• framework-default: generated defaults only;
• skipped: the operator explicitly skipped the interview.

Edit the file only when durable governing state has changed or the existing contract needs repair. Do not append per-run evidence review/history or mirror the final chat handoff; preserve durable decisions, unresolved conflicts, and provenance needed by future agents.