Skip to content

OpenClaw branch: align with Hermes main architecture (v0.13+ intent router) #32

Description

@scrimshawlife-ctrl

@prabu-openclaw — heads-up for the permanent openclaw branch.

We documented your OpenClaw Agent Skill port on Hermes main so both editions are discoverable, and we want the OpenClaw branch to match the Hermes architecture where practical (without losing OpenClaw packaging/state/doctor benefits you already landed).

Your work (credited)

Item Link
OpenClaw skill support 77fa721
Onboarding docs fix 15defae
Original PR #1
Permanent branch https://github.com/scrimshawlife-ctrl/Kubrick/tree/openclaw
Cross-link hub on main https://github.com/scrimshawlife-ctrl/Kubrick/blob/main/docs/OPENCLAW.md

Thanks for the Agent Skill packaging, external state dir (~/.openclaw/state/kubrick / KUBRICK_STATE_DIR), overlay evolution, doctor, and portability tests.

Hermes main architecture to match (target)

main has moved past the v0.8-era layout your branch forked from. Highest-value alignment for OpenClaw users:

1. Operator surface — intent router

2. Unified compiler / storyboard / multi-provider adapt

  • do compile pipeline (brief → retrieval → graph → storyboard → provider packet)
  • Providers: grok-imagine, flux, sd3, midjourney (syntax-only adapters)
  • Canonical example: examples/authority-transfer-storyboard/

3. Closed-loop visual QA + multi-signal learning

  • do visual --action closed-loop (geometry / state / residue / convergence reported separately)
  • do learn — forge-signals, outcome receipts, proposal-only evolution (no auto corpus mutation)
  • Schemas under schemas/ on main (forge-signal-bundle, multi-signal-evolution-receipt, etc.)

4. Governance invariants (keep on both editions)

  • Fail closed / NOT_COMPUTABLE on weak evidence
  • Local outputs stay PROPOSED / OBSERVATION
  • Evolution never auto-applies
  • Forge remains canonical when connected
  • Audience prompts free of named esoterica by default

What should stay OpenClaw-native (do not drop)

Please preserve when porting main features onto openclaw:

  • OpenClaw / ClawHub packaging (.clawhubignore, skill frontmatter appropriate for OpenClaw)
  • Install root ~/.openclaw/skills/kubrick
  • External mutable state (KUBRICK_STATE_DIR / ~/.openclaw/state/kubrick)
  • Overlay-based evolution (no in-package corpus mutation)
  • scripts/doctor.py + portability evals / CI on the OpenClaw track
  • No external symlink that OpenClaw would skip as untrusted

Suggested port approach

  1. Read docs/OPENCLAW.md on main (dual-edition contract).
  2. Rebase or cherry-pick from main onto openclaw in small slices: intent_router + kubrick.py first, then compile/adapters, then visual/learn schemas.
  3. Re-point OpenClaw docs/install so openclaw users get do <intent> as primary (same mental model as Hermes).
  4. Keep dual-track: do not silent-merge openclaw into main without review; port features deliberately both ways.

Docs already on main for you

Happy to answer questions on intent routing or the compiler pipeline if anything is ambiguous when you start the port.

cc maintainers of Hermes main — this issue is the handoff surface for edition alignment.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

enhancementNew feature or request

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions