ce-architecturelisted
Install: claude install-skill relusion/vg-sdlc-claude-plugins
# Architecture
**Invocation:** `$ARGUMENTS`
## Runtime Inputs
### Select one mode
Mode dispatch is the first action and performs no write.
| Input | Mode | Outcome |
|---|---|---|
| `explore:<draft-slug>` | Explore | Compare complete solution directions and run the human Architecture Direction workbench before decomposition. |
| `shape:<draft-slug>` | Shape | Test one parent-elected provisional plan candidate read-only and return its architecture impact. |
| `<plan-slug>` or no prefix | Baseline | Build and, after explicit human approval, publish the solution-architecture package for a written plan. |
For either prefix, require the complete input to match
`(?:explore|shape):[a-z0-9]+(?:-[a-z0-9]+)*`. Load only the matching companion
file and stop at its result:
- `explore` → `${CLAUDE_SKILL_DIR}/exploration-mode.md`. Its validated
`architecture-options.md` is the sole permitted domain write and requires
the deterministic `architecture-workbench.py` renderer plus an independent
`architecture-options-lint.py` pass before a selectable revision is shown.
The model authors semantic directions and reasoning only; it never
reverse-engineers validators or hand-builds report projections and hashes.
- `shape` → `${CLAUDE_SKILL_DIR}/shaping-mode.md`. It is repository-read-only.
A valid handoff proves `/core-engineering:ce-plan` already elected shaping;
do not add a nested scope or consent gate.
An input with neither prefix enters baseline mode. Accept only a canoni