Skip to content

feat: declarative deploy system (rbnx deploy/init/package-new) + new config schema - #28

Closed
lhw2002426 wants to merge 9 commits into
syswonder:devfrom
lhw2002426:add_deploy2
Closed

lhw2002426 wants to merge 9 commits into
syswonder:devfrom
lhw2002426:add_deploy2

Conversation

@lhw2002426

Copy link
Copy Markdown
Contributor

Summary

Introduces a declarative deployment workflow with new CLI commands and a restructured configuration system.

Changes

New Config Schema

  • robonix_manifest.yaml as the sole runtime entry point (system/primitives/services/skills)
  • Per-package config unified as package_manifest.yaml; build/start are required string fields supporting arbitrary commands (bash, make, binary, etc.)
  • Directory layout split by role: primitives/, services/, skills/
    New CLI Commands
  • rbnx deploy [-c config]: launches full stack in order (system → primitives → services → skills) with dependency-aware topological sorting
  • rbnx build --all [-c config]: batch build all packages in dependency order
  • rbnx init : scaffold a new project
  • rbnx package-new --type <primitive|service|skill>: scaffold a new package
    Pilot Improvement
  • VLM discovery now retries up to 30 times (2s interval) to tolerate late service startup
    Other
  • workspace.rs: topological sort (Kahn's algorithm), git clone for remote packages
  • Examples and test fixtures updated to new config format

@lhw2002426 lhw2002426 closed this Apr 26, 2026
Chenyantt pushed a commit to Chenyantt/robonix that referenced this pull request Aug 8, 2026
feat: add hot plugin for cap and aync mcp server
Chenyantt pushed a commit to Chenyantt/robonix that referenced this pull request Aug 8, 2026
The "system services" set is now: memory only. VLM moved into pilot
(commit fb90657), so the previous robonix/system/{vlm,memory}/ layout
collapses to just system/memory/.

Pure rename — code internals unchanged. The package still identifies as
com.robonix.example.memsearch_service and the inner Python module stays
`memsearch_service`. cap_id rename + actual code migration (service.py
refactor for the build-phase setup rule, switch to new Atlas API, etc.)
are tracked separately under tasks syswonder#28 / syswonder#36 — atlas's legacy shim
keeps the unmodified service running against the new Atlas in the
meantime, verified by the wire-compat smoke (3 MCP tools on shared
endpoint, register/declare/heartbeat/query/unregister all OK).

Untracked build artefacts (proto_gen/, robonix_mcp_types/, *.db, *.log,
agent_memory/) are left in their original location; they regenerate
into system/memory/ next time `rbnx codegen` runs.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant