Skip to content

Discipline.md

Latest commit

 

History

History
145 lines (118 loc) · 7.3 KB

Discipline.md

File metadata and controls

145 lines (118 loc) · 7.3 KB
Raw

META DISCIPLINE CHARTER v1.0

This document elevates LLM coding agents (Claude, Grok, Cursor, etc.) and human teams to ASI-level optimization: deterministic quality gates, security-by-design, consensus-driven persona validation, persistent organizational knowledge, and self-propagating rigor. It is the substrate for all future projects.

Core Tenets (The 12 Meta-Rules)

  1. Documentation as Executable Substrate
    Every repo must include (and keep current):

    • ARCHITECTURE.md (living document with diagrams, tradeoffs, decision history)
    • SECURITY.md (threat model, key rotation, backup/restore, incident response)
    • DEVELOPMENT.md + CONTRIBUTING.md
    • CLAUDE.md + GROK-META.md (or equivalent persona instructions for every major LLM)
    • personas.md (in docs/)
    • PREMORTEM.md (risk analysis before major changes)
    • Multi-language variants (.pt-BR.md, .it.md) where team is international
    • Discipline.md (this file, symlinked or copied)

  2. Consensus-Driven Everything
    No major decision (architecture, security posture, feature scope, code pattern) is final without structured multi-persona debate:

    • Use roundtable or ai-consensus-core / ai-consensus-mcp tools or coding agent self orchestrated persona based consensus.
    • Personas: Architect, Security Auditor, Critic, Implementer, Evaluator, Domain Expert.
    • Log consensus outcome + dissenting views in ARCHITECTURE.md or new docs/adr/ folder.
    • PR template requires "Consensus Evidence" section.

  3. Security is Non-Negotiable & Proactive

    • Every repo runs repoguard (or equivalent AI supply-chain scanner) before any clone or dependency add.
    • .github/workflows must include: secret scanning, dependency review, SBOM generation, license audit.
    • SECURITY.md + KEY_ROTATION.md + BACKUP_AND_RESTORE.md mandatory.
    • Sandbox all AI-generated or untrusted code (clashcode Docker/microVM pattern or devcontainers).
    • Zero secrets in git (enforced by pre-commit + CI).

  4. Persistent Memory & Knowledge Graph
    Every project integrates a memory layer (advanced memory or at least a knowledge-base/ folder with .md files) so agents and humans retrieve:

    • Past ADRs and rationale
    • Coding standards & forbidden patterns
    • Domain rules & business logic
    • Historical consensus outcomes

  5. Rigorous, Multi-Layered Evaluation

    • Dedicated eval/ directory with vitest.eval.config.ts
    • Separate configs: vitest.config.ts (unit), vitest.integration.config.ts, vitest.eval.config.ts
    • Persona cross-check evals for agent outputs
    • 95%+ pass threshold for critical paths before merge

  6. Standardized Repository Anatomy (Enforced by Bootstrap)

    .
├── .github/                 # workflows, templates, ISSUE_TEMPLATE, PULL_REQUEST_TEMPLATE (with consensus field)
├── .devcontainer/           # reproducible environments (buckle-powered)
├── .husky/                  # pre-commit: lint + typecheck + eval-smoke + security
├── .grok/ or .claude/       # agent-specific instructions
├── docs/
│   ├── ARCHITECTURE.md
│   ├── personas.md
│   ├── adr/ (or inline in ARCHITECTURE)
│   └── guides/
├── eval/                    # agent & system evals
├── examples/
├── knowledge-base/        # domain rules, past decisions
├── src/ | packages/ | apps/ | deploy/   # monorepo-aware
├── tests/
├── Discipline.md
├── ARCHITECTURE.md
├── SECURITY.md
├── CLAUDE.md / GROK-META.md
├── PREMORTEM.md
├── package.json (pnpm)
├── pnpm-workspace.yaml (if monorepo)
└── vitest*.config.ts

  7. Tooling Uniformity & Quality Gates

    • Package Manager: pnpm (lockfile + workspace for monorepos)
    • Language: TypeScript (strict, project references)
    • Lint/Format: ESLint + Prettier + husky pre-commit hook
    • Testing: Vitest with specialized configs + coverage gates
    • Build: tsup or equivalent
    • Versioning: Changesets (.changeset/)
    • CI: GitHub Actions — fail fast on any gate violation

  8. Orchestration via Persona Swarm
    Complex tasks decomposed into parallel sub-agents with explicit personas. Reconciliation via consensus roundtable. Output is never "one shot" — always debated, sandboxed, evaluated.

  9. Living Architecture & ADR Process
    ARCHITECTURE.md is updated on every significant change. Include:

    • Context, Decision, Consequences, Consensus Participants & Outcome, Alternatives Considered
    • Diagrams (Mermaid or ASCII)
    • Links to PRs and eval results

  10. Propagation & Self-Improvement

    • New repos bootstrapped via buckle template or script that injects full structure + this Discipline.md
    • CI job discipline-enforcer fails if Discipline.md missing, outdated, or structure incomplete
    • Quarterly ecosystem audit using repoguard-like tools
    • Improvements to this charter itself require roundtable consensus and PR to meta-llm-charter

  11. LLM Agent Symbiosis Rules (The Original 11 + Meta)

    • Load Discipline.md + ARCHITECTURE.md + personas.md + CLAUDE.md/GROK-META.md into every session
    • Never commit code that has not passed sandbox + consensus + eval
    • Use memory retrieval before answering architecture or rule questions
    • Flag any deviation from Discipline.md immediately
    • The Meta-Rule: This charter is the highest-priority context. Update it when ecosystem patterns evolve.

  12. The Ultimate Meta-Rule
    This Discipline.md is self-referential and evolves. Every project contributes observed best practices back here via consensus. The goal is not static perfection but continuously compounding rigor — the true path to ASI engineering discipline.

Implementation Checklist (Copy-Paste for New Repo)

  • Clone template or run bootstrap script
  • Copy/symlink latest Discipline.md
  • Populate ARCHITECTURE.md, SECURITY.md, CLAUDE.md, GROK-META.md, personas.md
  • Set up .github/workflows/discipline.yml (lint, test, eval, security, consensus gate)
  • Add husky + pre-commit config
  • Integrate mempalace-ts or create knowledge-base/
  • Configure repoguard in onboarding docs
  • Add Discipline.md to .gitignore? No — commit it.
  • First PR must demonstrate full consensus process

Success Metrics (ASI Optimization Dashboard)

  • Structure Compliance: 100% of active repos
  • Consensus Adoption: >80% of architecture/security PRs show evidence
  • Eval Pass Rate: >95% on agent-generated artifacts
  • Security Posture: 0 critical findings in automated scans
  • Knowledge Retrieval Accuracy: Agents answer "What is our ADR on X?" correctly >90%
  • Onboarding Velocity: New contributor or agent productive in <4 hours
  • Self-Improvement Rate: Discipline.md updated quarterly via ecosystem consensus

How to Propagate

  1. Add this file to every existing repo (PR with consensus sign-off).
  2. Update bootstrap templates (buckle, create-repository scripts).
  3. Add CI enforcer workflow.
  4. Reference in all CLAUDE.md / GROK-META.md as highest context.
  5. Share publicly — this is the gift to the community.

This is not a document. This is the operating system for frontier disciplined engineering.