split-architecture

name	split-architecture
description	This skill should be used when converting a single-file architecture doc to directory-based format for better organization of large architecture documents

Split Architecture Skill

Converts a single-file architecture document ({{specs_dir}}/architecture.md) into a directory-based format ({{specs_dir}}/architecture/) for better organization and targeted context loading.

When to Use

• Automatically by design-architecture and source-architecture-from-code when the single-file architecture doc crosses 500 lines or 10 decision records (sections matching ### DR-\d+)
• Manually when the user invokes /split-architecture

Step 0: Resolve Project Context

1. Monorepo check: Does .groundwork.yml exist at the repo root?
   • If yes → Is {{project_name}} non-empty?
     • If empty → Invoke Skill(skill="groundwork:select-project") to select a project, then restart this skill.
     • If set → Project is {{project_name}}, specs at {{specs_dir}}/.
   • If no → Continue (single-project repo).
2. Proceed with the resolved project context.

Step 1: Validate Preconditions

1. Check single file exists: {{specs_dir}}/architecture.md must exist
   • If missing → output "No single-file architecture doc found at {{specs_dir}}/architecture.md. Nothing to split." and stop.
2. Check directory does NOT exist: {{specs_dir}}/architecture/ must not already exist
   • If it exists → output "Architecture doc is already in directory format at {{specs_dir}}/architecture/. Nothing to do." and stop.
3. Read the file and count lines and decision records (sections matching ^### DR-\d+).

Step 2: Split the File

Parse the single-file architecture doc into sections and write each to its own file under {{specs_dir}}/architecture/.

Directory Structure

{{specs_dir}}/architecture/
├── _index.md                 # Header, metadata
├── 01-executive-summary.md   # Section 1: Executive Summary
├── 02-architectural-drivers.md # Section 2: Architectural Drivers
├── 03-system-overview.md     # Section 3: System Overview
├── 04-components/            # Section 4: One file per component
│   ├── _index.md             # Section 4 header ("## 4) Component Details")
│   └── <component-slug>.md   # e.g., cli.md, query-router.md
├── 05-data.md                # Section 5: Data Architecture
├── 06-integration.md         # Section 6: Integration Architecture
├── 07-security.md            # Section 7: Security Architecture
├── 08-infrastructure.md      # Section 8: Infrastructure & Deployment
├── 09-observability.md       # Section 9: Observability
├── 10-cost-model.md          # Section 10: Cost Model
├── 11-decisions/             # Section 11: One file per DR
│   ├── _index.md             # Section 11 header ("## 11) Decision Records")
│   └── DR-NNN.md             # e.g., DR-001.md, DR-015.md
├── 12-open-questions.md      # Section 12: Open Questions & Risks
└── 13-appendices.md          # Section 13: Appendices

Parsing Rules

1. _index.md — Everything from the start of the file up to (but not including) ## 1). This includes the header block (Version, Last updated, Status, Owner).

2. 01-executive-summary.md — From ## 1) up to (but not including) ## 2).

3. 02-architectural-drivers.md — From ## 2) up to (but not including) ## 3).

4. 03-system-overview.md — From ## 3) up to (but not including) ## 4).

5. 04-components/_index.md — The ## 4) heading line and any text before the first ### 4. subsection. Typically just:

   ## 4) Component Details
   

6. 04-components/<component-slug>.md — One file per ### 4.N subsection. The filename is a slug of the component name:

   • ### 4.1 CLI → cli.md
   • ### 4.2 Query Router → query-router.md
   • ### 4.15 MCP Server → mcp-server.md
   • Preserve the full ### 4.N ... heading in the file

7. 05-data.md — From ## 5) up to (but not including) ## 6).

8. 06-integration.md — From ## 6) up to (but not including) ## 7).

9. 07-security.md — From ## 7) up to (but not including) ## 8).

10. 08-infrastructure.md — From ## 8) up to (but not including) ## 9).

11. 09-observability.md — From ## 9) up to (but not including) ## 10).

12. 10-cost-model.md — From ## 10) up to (but not including) ## 11).

13. 11-decisions/_index.md — The ## 11) heading line and any text before the first ### DR- subsection.

14. 11-decisions/DR-NNN.md — One file per ### DR-NNN subsection. The filename matches the DR number:

    • ### DR-001: Project Structure → DR-001.md
    • ### DR-015: Call Graph Data Model [V3] → DR-015.md
    • Preserve the full ### DR-NNN: ... heading in the file

15. 12-open-questions.md — From ## 12) up to (but not including) ## 13).

16. 13-appendices.md — From ## 13) to the end of the file.

Edge Cases

• If a section is missing from the source file, skip that output file (don't create empty files).
• If non-standard numbered sections exist beyond 13, include them as NN-slug.md files.
• Preserve all horizontal rules (---) as they appear in the source.
• Preserve trailing newlines — each output file should end with a single newline.

Step 3: Delete the Single File

After all directory files are written and verified:

1. Verify — Read back _index.md, at least one component file, and at least one DR file to confirm they look correct.
2. Delete — Remove {{specs_dir}}/architecture.md using rm.
3. Confirm — Output a summary:

Split architecture.md into directory format:
  {{specs_dir}}/architecture/
  ├── _index.md
  ├── 01-executive-summary.md
  ├── 02-architectural-drivers.md
  ├── 03-system-overview.md
  ├── 04-components/ (N component files)
  ├── 05-data.md
  ├── ...
  ├── 11-decisions/ (N decision records)
  ├── 12-open-questions.md
  └── 13-appendices.md

The single file has been removed. All skills (design-architecture, source-architecture-from-code, etc.) already support directory mode.

Auto-Split Thresholds

When invoked automatically (not by the user), the caller has already verified that at least one threshold is crossed:

• Line count: wc -l on {{specs_dir}}/architecture.md >= 500
• Decision record count: Number of ### DR-\d+ headings >= 10

No user confirmation is needed — the split happens silently and the caller reports what happened.

When invoked manually by the user, skip threshold checks and split unconditionally.