Skip to content

Governances

Governances are versioned, agent-readable rule sets that encode what to do — imperative rules, conventions, and standards that agents load on demand to complete a workflow correctly.

They answer the question the agent has right now: “what are the rules for this?” ADRs answer a different question: “why did we decide this?” Agents read governances. Humans read ADRs.

A governance is a dense, self-contained document:

  • Imperative rules — must, should, do not; no tutorials or background
  • No rationale sections — no ## Why or ## Background; that belongs in ADRs
  • Self-contained — agent completes the workflow from the governance alone
  • References at end — only links to other governances or external docs

Agents load governances via CLI, not by reading files directly:

Terminal window
npx cyberplace@<version> governance show skill-design
npx cyberplace@<version> governance show skill-design --format agent

Pinning the version ensures agents always get rules that match the installed tooling.

NamePurpose
skill-designRules for authoring SKILL.md files
skill-repo-structureRules for organizing a skill library repository
agent-tool-outputOutput rules for scripts, hooks, and CLIs that agents invoke
cli-resolutionStrategy for invoking a Node CLI that may be global, repo-local, or absent
universal-pluginFormat spec for plugins that work across multiple agent harnesses

Both governance and discipline are verified the same way: give the agent a scenario, observe what it does, check the outcome. The distinction is not how you test compliance — it is:

GovernanceDiscipline
When activeOn demand, per workflowAlways-on, any agent or sub
Content shapeNormative rules for a specific domainCross-cutting behavioral habits
Static analysisYes — tools like audit-skill check artifacts against governance rulesNo — only verifiable by observing behavior

Tagline: Governance defines what is correct. Discipline defines what is habitual.