SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmb-init

mb-init

Agents & models

View repository

Updated 10-16 06:49

About this skill

mb-init — Initialize Memory Bank skeleton
  • What it does: creates the base folders, core docs, and project-native command proxies.
  • Use it when: you want the skeleton only, without immediately choosing PRD or brownfield mapping.
  • Input: repository root.
  • Output: .memory-bank/, .tasks/, .protocols/, AGENTS.md, and generated command stubs.

Goal

Create a consistent baseline so agents can work with repo-local context instead of ad-hoc prompts.

Steps

1) Create directories

Create (if missing):

  • .memory-bank/ with subfolders (see ./references/shared-structure-template.md)
  • .tasks/
  • .protocols/

2) Create core Memory Bank files

From templates, create:

  • AGENTS.md
  • CLAUDE.md → symlink/copy to AGENTS.md
  • .memory-bank/index.md
  • .memory-bank/constitution.md
  • .memory-bank/mbb/index.md
  • .memory-bank/spec-index.md
  • .memory-bank/glossary.md
  • .memory-bank/invariants.md
  • .memory-bank/product.md
  • .memory-bank/requirements.md
  • .memory-bank/schemas/task.schema.json
  • .memory-bank/tasks/index.json
  • .memory-bank/testing/index.md
  • .memory-bank/skills/index.md

Fresh PRD-less bootstrap must not create .memory-bank/features/FT-001-*.md, .memory-bank/tasks/TASK-001.task.json, or any other fake roadmap artifact. .memory-bank/tasks/index.json starts as { "version": 1, "tasks": [] }; /write-prd creates a clarified PRD, /spec-init prepares only the lightweight SDD route map, /prd creates real features, mandatory /spec-design creates a minimal or full backbone gate, /spec-improve FT-<NNN> completes or marks unnecessary feature design, and /prd-to-tasks FT-<NNN> creates real TASK-*.task.json records after that. /clarify-feature FT-<NNN> is optional for explicitly pending/blocked features.

Also create optional folders that support the richer normative layer without making it mandatory:

  • .memory-bank/contracts/
  • .memory-bank/states/
  • .memory-bank/runbooks/

3) Create command stubs

So links from AGENTS.md are not broken, create minimal docs under .memory-bank/commands/. Use ./references/shared-commands-*.md:

  • mb.md
  • analysis.md
  • brainstorm.md
  • brief.md
  • constitution.md
  • write-prd.md
  • spec-init.md
  • prd.md
  • spec-design.md
  • spec-improve.md
  • spec-auto.md
  • clarify-feature.md
  • prd-to-tasks.md
  • execute.md
  • verify.md
  • red-verify.md
  • autopilot.md
  • autonomous.md
  • map-codebase.md
  • discuss.md
  • add-tests.md
  • review.md
  • find-skills.md

4) Create native skills (proxy commands)

Create thin proxy skills for each command so they work natively across runtimes:

  • .claude/skills/<name>/SKILL.md → Claude Code + OpenCode
  • .agents/skills/<name>/SKILL.md → Codex CLI + OpenCode

Each proxy contains: Read and follow the instructions in .memory-bank/commands/<name>.md. This registers commands natively: /mb in Claude Code, $mb in Codex, /mb in OpenCode.

The init-mb.js script creates both sets automatically.

Agents must read .memory-bank/constitution.md early during priming. It records project governing principles and does not replace .memory-bank/invariants.md, .memory-bank/contracts/*, or .memory-bank/spec-index.md; use /constitution only to create or amend those principles.

5) Enforce MBB rules immediately

  • Every .memory-bank/**/*.md must have frontmatter (description, status).

Optional fast path

If Node.js is available, you may run the helper script (safe: doesn’t overwrite existing files):

# Option A: copy ./scripts/shared-init-mb.js into your repo then run:
node scripts/init-mb.js

If you don’t want a script, just create the files manually using the templates.

Definition of done

  • AGENTS.md exists and points to .memory-bank/index.md.
  • CLAUDE.md exists and mirrors AGENTS.md.
  • .memory-bank/ has the seeded docs.
  • .memory-bank/constitution.md exists as the governing-principles doc and /constitution is available in .memory-bank/commands/.
  • .memory-bank/tasks/index.json and .memory-bank/schemas/task.schema.json exist; task state is JSON-backed.
  • .memory-bank/tasks/index.json has an empty tasks array in a PRD-less skeleton.
  • Skeleton bootstrap creates no fake feature docs; task planning starts later with /write-prd, /spec-init, /prd, mandatory /spec-design, /spec-improve FT-<NNN>, then /prd-to-tasks FT-<NNN>.
  • No .memory-bank/tasks/TASK-001.task.json is created by bootstrap.
  • Future task records must contain mandatory tier: T0|T1|T2|T3; routing is only through task.tier, not the removed risk / risk.level model.
  • .memory-bank/commands/ has stub command docs.
  • .tasks/ and .protocols/ exist.