Skip to content

docs(skills): document ZCode's executed-vs-recorded component boundary#28

Merged
rldyourmnd merged 1 commit into
mainfrom
docs/zcode-execution-boundary
Jul 18, 2026
Merged

docs(skills): document ZCode's executed-vs-recorded component boundary#28
rldyourmnd merged 1 commit into
mainfrom
docs/zcode-execution-boundary

Conversation

@rldyourmnd

Copy link
Copy Markdown
Contributor

What

Documents which ZCode 3.3.6 plugin-manifest fields actually execute, and which are recorded but inert — so the toolkit is honest about its coverage and no agent authors dead config.

Why (the LSP question, settled)

ZCode's own built-in guide (zcode-guidezcode-configuration-guide, diagnosing-plugins) states, twice:

Component fields: commands, skills, hooks, mcpServers, agents … The fields channels, lspServers, outputStyles, and settings are recorded but not executed.

So there is no loadable LSP component on this runtime. An add-lsp skill would teach a non-feature — the same "installed but not loading" trap already fixed for skills. The correct fix is documentation, not a new skill. This means the marketplace already covers 100% of the executable extension surface (skills, commands, agents, hooks, MCP servers).

Changes

  • references/zcode-native-format.md: new Component execution model section (executed vs recorded-but-not-executed, with the v2-setting vs inert-settings distinction called out); fixed a stale skill name (repo-orientationnddev-builder-orientation).
  • nddev-builder-orientation: new "Fields ZCode records but never executes" subsection.
  • add-plugin: manifest guardrail extended to the four inert fields.
  • Locked in by test_execution_boundary_is_documented (harness side, same PR wave).

Scope

Prose only — no component, command, agent, path, or runtime behavior changed. release-evidence.json digest refreshed (pending, bound to a stable harness origin/main ancestor).

Validation

  • test_marketplace_structure.py: 13 passed (incl. the new boundary test)
  • validate_zcode_components.py / validate_skills.py / markdown fences: PASS
  • validate_public_contracts.py: PASS
  • emit_evidence.py --check: valid (pending)

https://claude.ai/code/session_01Y29c9sJRhChvxDcUjab5Xo

ZCode 3.3.6 executes only the commands/skills/hooks/mcpServers plugin-manifest
fields (plus agents, which load from user scope via the installer flatten). Its
own built-in guide (zcode-configuration-guide, diagnosing-plugins) states that
`channels`, `lspServers`, `outputStyles`, and `settings` are recorded but never
executed.

Authoring any of those yields dead config. In particular there is no loadable
LSP component on this runtime, so an `add-lsp` skill would teach a non-feature —
the same "installed but not loading" trap already fixed for skills. Document the
boundary instead:

- references/zcode-native-format.md: add a "Component execution model" section;
  fix a stale skill name (repo-orientation -> nddev-builder-orientation).
- nddev-builder-orientation: add "Fields ZCode records but never executes".
- add-plugin: extend the manifest guardrail to the four inert fields.

Locked in by test_execution_boundary_is_documented (harness side). Prose only;
no component, command, agent, or runtime behavior changed. Refresh the
release-evidence digest.

Claude-Session: https://claude.ai/code/session_01Y29c9sJRhChvxDcUjab5Xo
@rldyourmnd
rldyourmnd merged commit c111cde into main Jul 18, 2026
8 checks passed
@rldyourmnd
rldyourmnd deleted the docs/zcode-execution-boundary branch July 18, 2026 06:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant