monadvault-capability-packager

Create, audit, improve, or package repeated Steward workflows as focused MonadVault capabilities or Codex Skills. Use when a Steward asks to create a skill, convert a workflow into a capability, impro

SKILL.md

GitHub

MonadVault Capability Packager

Overview

Use this skill to turn a repeated Steward workflow into a portable capability package. It helps decide what can live as a Skill and what must remain runtime infrastructure.

Routing

  • Use this skill for reusable capability boundaries, packaging, metadata, and validation design.
  • Use monadvault-found-monad when the requested artifact is a complete governed Monad rather than a reusable workflow.
  • Use monadvault-privacy-sentinel to audit the proposed scopes and exposure paths.

Conversion Decision

Convert to a Skill when the workflow is mostly:

  • procedural guidance
  • domain knowledge
  • prompt structure
  • review checklist
  • repeatable synthesis
  • document or memory shaping
  • capability packaging

Keep in MonadVault runtime when the workflow requires:

  • accounts, sessions, or auth
  • persistent encrypted storage
  • billing, ledgers, or provider keys
  • publication access control
  • moderation state
  • database writes
  • visitor entitlements
  • live hosted invocation

Packaging Workflow

  1. Name the capability with lowercase hyphen-case.
  2. Write a trigger-rich description that says exactly when to use it.
  3. Define the primary output contract.
  4. Choose resources:
    • references/ for templates, checklists, schemas, and domain notes.
    • scripts/ only for deterministic utilities.
    • assets/ only for reusable output files.
  5. Map required scopes conservatively:
    • prefer memory:read_selected
    • avoid broad memory or publication scopes
    • block billing:write, vault:export, publication:publish, and other high-risk scopes unless runtime governance exists
  6. Add review gates for privacy, publication, provider exposure, and monetization.
  7. Validate that the skill works without hidden runtime privileges.

Output Contract

Return:

  • Skill name
  • Description
  • Use cases
  • Non-goals
  • Required scopes
  • Install mode
  • SKILL.md outline
  • References/assets/scripts needed
  • Safety review
  • Test prompts

Use capability-spec-template.md for a package spec.

Guardrails

  • Do not pretend a Skill can replace runtime controls.
  • Do not request broad scopes because they might be useful later.
  • Keep the first version small enough to trigger reliably.
  • Prefer one focused skill over a vague mega-skill.