sdd reverse

Extracts SDD specs from legacy codebases.

Overview

sdd reverse analyzes existing code and automatically generates SDD spec drafts. Useful when introducing SDD to legacy projects.

Workflow

scan → extract → review → finalize

Subcommands

scan

Scans and analyzes project structure.

sdd reverse scan                    # Scan entire project
sdd reverse scan src/               # Scan specific directory only
sdd reverse scan --json             # Output as JSON

Output:

• Directory structure
• Language distribution statistics
• Estimated domain list
• File/symbol counts

extract

Extracts spec drafts from code based on scan results.

sdd reverse extract                 # Extract all
sdd reverse extract --domain auth   # Extract specific domain only
sdd reverse extract --depth deep    # Deep analysis (includes scenarios)
sdd reverse extract --dry-run       # Preview only

Options:

Option	Description
--domain <name>	Extract specific domain only
--depth <level>	Analysis depth (shallow, medium, deep)
--min-confidence <n>	Minimum confidence (0-100)
--dry-run	Preview without creating files

review

Reviews and approves/rejects extracted spec drafts.

sdd reverse review                  # Show pending review list
sdd reverse review auth/login       # View specific spec details

Review statuses:

• pending: Awaiting review
• approved: Approved
• rejected: Rejected
• needs_revision: Needs revision

finalize

Converts approved specs to official SDD specs.

sdd reverse finalize --all          # Finalize all approved specs
sdd reverse finalize auth/login     # Finalize specific spec only

Output Files

File	Description	Created By
.sdd/.reverse-meta.json	Scan result metadata	scan
.sdd/.reverse-drafts/<domain>/<spec>.json	Spec drafts	extract
.sdd/specs/<domain>/<spec>/spec.md	Finalized specs (v1.3.0)	finalize

v1.3.0 Path Change

Finalized specs are now saved in <domain>/<feature>/spec.md format. Example: .sdd/specs/auth/login/spec.md

Usage Examples

Basic Workflow

# 1. Scan project
sdd reverse scan
# → Directory structure, language distribution, domain estimation

# 2. Extract specs
sdd reverse extract --depth deep
# → Drafts created in .sdd/.reverse-drafts/

# 3. Review
sdd reverse review
# → Approve/reject/revise each spec

# 4. Finalize
sdd reverse finalize --all
# → Official specs created in .sdd/specs/<domain>/<feature>/spec.md

Extract Specific Domain Only

sdd reverse scan src/auth/
sdd reverse extract --domain auth
sdd reverse review auth/login
sdd reverse finalize auth/login

Confidence Scores

Extracted specs include confidence scores:

Factor	Weight	Description
documentation	25%	JSDoc/comment quality
naming	20%	Naming convention compliance
structure	20%	Code structure quality
testCoverage	20%	Estimated test coverage
typing	15%	Type information quality

Serena MCP Integration

When Serena MCP is connected, precise symbol-level analysis is available:

• Class/function/interface extraction
• Reference relationship analysis
• Dependency graph generation

Basic file scanning works even without Serena.

References

• Reverse Spec Extraction Guide
• Domain Management