Skip to content
AletheIA Docs
Esc
navigateopen⌘Jpreview
On this page

Roles, runtimes, and Adaptive Skills

Keep portable responsibilities, execution environments, and reusable methods distinct when operating AletheIA.

Three ideas are often mixed together in AI-assisted work. AletheIA keeps them deliberately separate:

Why the distinction matters

When these concepts collapse into one label, teams lose clarity about ownership:

If they are mixed What becomes unclear AletheIA response
“Codex is the reviewer” Is the responsibility portable, or tied to one product? Record reviewer as the role and Codex as the runtime.
“Use the testing agent” Is testing a role, a method, or an approval? Keep the role and chosen skill explicit.
“The PM agent approved it” Was there a human decision, a persona, or a runtime output? Treat personas as illustrative unless a contract defines a real role.
“The runtime has guardrails” Which boundary is enforced versus advisory? Identify the actual local permission or human gate.

Canonical portable roles

Not every Work Slice needs all five roles. Use the smallest set that makes the boundary, decision, and proof clear.

Example: one documentation correction

The same example can run in different environments:

Layer One possible choice What remains stable
Role implementer Bounded execution responsibility
Runtime Codex Local tools and session mechanics
Skill testing Verification method, when it materially shapes the work
Governance AletheIA Work Slice Scope, evidence, stop line, and closure decision

Runtime adapters

Where Adaptive Skills fits

Adaptive Skills provides portable methods that may support a role—for example, planning, debugging, testing, research, or documentation. A skill is selected because its method is useful for the slice, not because it grants authority.

Work Slice governance (AletheIA)
  └── active role (portable responsibility)
        └── selected skill (reusable method, if useful)
              └── runtime adapter (local execution mechanics)

The skill should appear in a handoff or record only when it materially influenced the method, validation, or continuation path.

Non-goals

This model does not:

  • introduce more canonical roles;
  • turn presentation personas into available agents;
  • require every runtime to have native subagents;
  • turn the role catalog into an automatic router;
  • change Adaptive Skills behavior, capability metadata, or governance;
  • authorize a runtime, tool, or security-sensitive action.

Next steps

Was this page helpful?