Create or update technical architecture after Proposal Review and before the specification is finalized.
Scanned 9/2/2026
Install to Claude Code
npx -y skills add xiongxianfei/rigorloop --skill architecture --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Architecture?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/xiongxianfei-architecture)More formats (shields.io, HTML) on the badges page.
---
name: architecture
version: "1.0.0"
schema-version: skill-readability-v1
description: >
Create or update technical architecture after Proposal Review and before the specification is finalized.
argument-hint: [feature spec path, proposal path, architecture question, or change name]
---
# Architecture authoring
## Workflow role
- role_name: architecture
- stage: authoring
- upstream: accepted proposal and embedded feasibility evidence
- downstream: spec reconciliation, then design-review
- summary: Author the smallest architecture package that establishes the technical design envelope.
- ownership: Architecture artifacts and authorized architecture authoring evidence only.
- must_not_claim: Design Review approval, plan readiness, implementation readiness, verification, branch readiness, release readiness, or PR readiness.
Portable work is isolated. Workflow-managed execution does not enlarge the architecture write set.
## Evidence and upstream settlement
Read current guidance, the accepted proposal and feasibility evidence, draft specification when present, canonical architecture, relevant ADRs, project map, and affected system surfaces. Rank governing artifacts above history and inference; upstream artifacts and state are read-only.
Require an accepted current proposal and feasibility evidence without later contradiction or open resolution. Route unsettled direction to `proposal` and behavior needing reconciliation to `spec`.
## Classification and loaded assemblies
Classify before mutation. Authoring mode is `portable` or `governed`; authoring action is `canonical-update`, `adr-only`, `canonical-update-with-adr`, `no-change-required`, or `blocked`.
| Assembly | Use | Loaded procedure |
| --- | --- | --- |
| `AA1-portable-authoring` | portable architecture authoring | `SKILL.md` plus package method |
| `AA2-governed-authoring` | governed authoring | AA1 plus governed authoring |
Loading selects procedure, not authority.
## Scope and routing
Treat specification requirements as SRs and architecture as their technical realization, not as another requirement level.
Use architecture for cross-component structure, data flow, persistence, APIs, deployment, packaging, adapters, security, quality targets, cross-cutting rules, or durable decisions. Use the smallest surface: canonical truth, ADR decision history, or both.
For a focused change with no new architectural decision, confirm the current architecture envelope with `no-change-required`; the current architecture remains the package member. Never create temporary architecture to resolve product or behavioral uncertainty.
Portable authoring repeats current applicability, writes only resolved architecture or ADR files, and never writes lifecycle, review, routing, or automation state.
## Governed signals and targets
Governed signals are `no-governed-signal`, `single-governed-candidate`, or `invalid-or-ambiguous-governed-signal`. Any explicit change ID, workflow identity, owning-change field, or governed entry counts even when malformed. Only no signal permits portable authoring; invalid, stale, conflicting, duplicated, escaped, unsafe, or ambiguous signals stop without portable fallback.
Target operations are `create`, `revise`, `supersede`, and `deprecate`. Canonical architecture and each ADR are distinct manifest targets. Never rewrite accepted history as though a prior decision never existed.
## Universal write and handoff boundaries
Author only accepted design required by the spec. Exclude milestones, secrets, review settlement, unsupported claims, and mutable state. Preserve history and identify replacement or supersession.
Governed authoring ends each completed target at `review-required` and hands off to specification reconciliation, then `design-review`. It never approves ADRs, settles supersession, or advances workflow. Report partial targets and blockers.
## Stop conditions and claims
Stop on missing or unreadable required resources, unresolved placement, invalid signals, stale basis, illegal state, ambiguity, conflicts, changed baseline, unsafe dependencies, unrecorded files, concurrency, incomplete recovery, or placeholders. The common path must not reconstruct missing procedure.
Never claim Design Review approval, ADR settlement, plan or implementation readiness, validation, verification, branch readiness, release, deployment, publication, or PR readiness.
## Resource map
- READ `references/requirement-to-delivery-model.md` when relating system requirements to technical realization or downstream allocation.
- READ `references/architecture-package-method.md` for `AA1-portable-authoring` and `AA2-governed-authoring` before package judgment or writes.
- READ `references/governed-architecture-authoring.md` only for `AA2-governed-authoring`; validate exact authority before governed reads or writes.
- COPY `assets/architecture-skeleton.md` when creating or fully rewriting canonical architecture. Fill applicable sections and remove placeholders.
- COPY `assets/adr-skeleton.md` when creating a new ADR. Fill every applicable field.
- COPY `assets/diagram-styles.mmd` when Mermaid flowchart or graph diagrams need the standard role styles.
Require readable, contained, same-version resources. Missing triggered resources stop dependent work.
## Evidence collection efficiency
Use summary and stable-ID first reasoning. Prefer check IDs, requirement IDs, file paths, line citations, diffs, and targeted excerpts.
## When full-file read is required
Read fully when the whole file is the review target, bounded searches disagree, isolation is unsafe, or a behavior-changing edit depends on the whole source-of-truth artifact.
## Output skeleton
```md
COPY `<applicable mapped asset>` to `<resolved target>` and replace every placeholder.
```
Use mapped assets only for applicable authoring output; `no-change-required` uses the compact result below.
## Expected output
Return a concise result without unfilled placeholders.
## Result
Report authoring mode, action, assembly, targets, changed sections or ADRs, blockers, recording state, claim limitations, and next stage.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!