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

Runtime Adapter — Claude Code

Goal

Describe how a project can map AletheIA’s portable agent roles onto Claude Code while keeping Claude-local agent mechanics as adapter detail rather than framework truth.


Local runtime shape

Claude Code is often used with local agent definitions, role presets, or runtime-local agent files.

Typical properties:

  • agent structure may be more declarative than tool-based runtimes
  • local agent files or metadata may carry runtime-specific configuration
  • the same canonical role may be implemented through richer runtime-local conventions

This is healthy as long as the framework meaning stays portable.


Core adapter rule

Claude Code may use runtime-local agent files, frontmatter, or agent-specific metadata.

Those are adapter concerns. They should not silently replace the canonical role contract.

A healthy Claude Code mapping still preserves three things separately:

  1. canonical role
  2. Claude-local agent definition or class
  3. selected skills

Canonical role Healthy Claude Code mapping Notes
orchestrator main Claude Code working boundary should own framing, delegation choice, and integration
explorer Claude-local research or inspection agent should reduce discoverable unknowns without absorbing implementation
implementer Claude-local execution agent can be more declarative locally, but still owes bounded outputs
reviewer Claude-local review or critique agent useful when the runtime has strong review-oriented agent patterns
validator Claude-local validation or QA-oriented agent should stay attached to evidence, not broad reinterpretation

Claude Code may expose more runtime-local structure than Codex. That does not change the canonical role model.


Skills in the Claude Code adapter

Claude Code may pair portable roles with the same skills library used in other runtimes.

That means a team can preserve continuity such as:

  • implementer + testing + debugging
  • reviewer + architecture-review + communication
  • validator + testing + workflow

The role still defines:

  • mission
  • stop rules
  • handoff triggers
  • proof expectation

The skill only improves the execution discipline inside that role.


Runtime-local richness rule

Claude Code may support richer local agent metadata than other runtimes.

Use that richness when it helps locally, but keep these distinctions explicit:

  • local metadata is not the canonical role definition
  • local file format is not framework truth
  • local agent naming may differ from the portable role catalog

When in doubt, preserve the portable role in the handoff and treat the Claude-local configuration as an implementation detail.


Handoff pattern

A Claude Code handoff should preserve at least:

  • current canonical role
  • receiving canonical role when relevant
  • Claude-local agent reference if it matters for continuation
  • selected skills when they materially shape the work
  • proof already obtained
  • next action

If the next boundary moves to Codex or another runtime, keep the canonical role stable whenever the semantic responsibility did not change.


Non-goals

This adapter should not assume:

  • Claude-local agent files define the framework core
  • every local preset must become a new canonical role
  • runtime richness justifies collapsing role and skill into one artifact

The adapter can be rich locally while the framework stays portable.

Was this page helpful?