Professional Software Specification Generator for Claude Code

Generate industry-standard specifications — from early-stage brainstorming to Technical Design with auto-generated feature specs — each usable standalone or as part of a streamlined auto chain.

Software projects need clear specifications. spec-forge covers the full journey from idea to implementation-ready documents:

Aliases: /prd, /srs, /tech-design, /test-cases, /idea, /decompose, /review, /audit, /analyze work as shortcuts — they invoke each skill directly, bypassing the /spec-forge orchestrator.

• Idea to Spec: Brainstorm interactively, then graduate ideas into architecture docs + feature specs
• Full Chain Mode: One command runs the streamlined chain (Idea → Decompose → Tech Design + Feature Specs → Review)
• Standalone or Chained: Use any command on its own, or run the full chain for traceability
• Industry Standards: Templates grounded in Google, Amazon, Stripe, IEEE, and ISTQB best practices
• Automatic Context Scanning: Scans your project structure, README, and existing docs before generation
• Project Decomposition: Automatically analyzes scope and splits large projects into sub-features
• Smart Upstream Detection: Finds upstream documents when available; asks compensating questions when not
• Quality Checklists: Built-in 4-tier validation (completeness, quality, consistency, formatting)
• Mermaid Diagrams: Architecture, sequence, user journey, and Gantt diagrams
• Spec Review: Review generated specs for completeness, consistency, and actionability with auto-fix
• Documentation Audit: Cross-reference docs against code for quality, completeness, and consistency
• Document Landscape Analysis: Map, cluster, and evaluate document ecosystems

Interactive, multi-session brainstorming for early-stage ideas:

• Iterative: Explore an idea across multiple sessions, days apart
• Persistent: Sessions stored in ideas/ directory (add to .gitignore or commit for team use)
• Graduated: When an idea is ready, it flows into the spec chain seamlessly

/spec-forge:idea cool-feature       # Start or resume brainstorming
/spec-forge:idea                    # List all ideas

Status flow: exploring → researching → refining → ready → graduated (or parked at any stage)

Run the streamlined specification chain in one command:

/spec-forge user-login              # Auto: Idea → Decompose → Tech Design + Feature Specs

• Detects existing documents and resumes from where you left off
• Idea stage is interactive; Tech Design minimizes questions when idea draft exists
• Auto-generates per-component feature specs in docs/features/ (Step 7 of tech-design)
• If an idea draft exists in ideas/, uses it as requirements context

Analyze project scope and split into sub-features if needed:

/spec-forge:decompose my-project     # Interview → split analysis → manifest

• Lightweight 3-5 round interview focused on scope boundaries
• Generates docs/project-{name}.md manifest for multi-split projects
• Automatically invoked as Stage 2 when running /spec-forge <name> full chain (after Idea, before Tech Design)

Generates a Product Requirements Document including:

• Problem statement and product vision
• User personas and user stories
• Feature requirements with P0/P1/P2 prioritization
• Success metrics (KPI/OKR)
• User journey maps (Mermaid)
• Timeline and milestones (Mermaid Gantt)
• Risk assessment matrix

Reference: Google PRD, Amazon Working Backwards (PR/FAQ), Stripe Product Spec

Generates a Software Requirements Specification including:

• Functional requirements with structured IDs (FR-XXX-NNN)
• Non-functional requirements (NFR-XXX-NNN)
• Data model and data dictionary
• External interface requirements
• Requirements traceability matrix (PRD → SRS, when PRD exists)

Standalone: When no upstream PRD is found, asks additional questions to compensate.

Reference: IEEE 830, ISO/IEC/IEEE 29148, Amazon Technical Specifications

Generates a Technical Design Document including:

• C4 architecture diagrams (Context, Container, Component)
• Alternative solution comparison matrix
• API design (RESTful / GraphQL / gRPC)
• Database schema and migration strategy
• Security, performance, and observability design
• Deployment and rollback strategy

Standalone: When no upstream PRD/SRS is found, asks additional questions to compensate.

Reference: Google Design Doc, RFC Template, Uber/Meta Engineering Standards

Generates structured test cases with multi-dimensional coverage:

• Auto-scans project to extract testable units (APIs, functions, components, CLI commands, tools)
• Detects project profile (Web API, CLI, Frontend, AI Agent, etc.) and adapts output
• Multi-dimensional coverage: L1 (Happy Path) + L2 (Boundary/Error) + L3 (Negative)
• Coverage matrix with gap analysis
• Test strategy and methodology (default); --formal adds management sections (environment, roles, schedule)
• Downstream integration: output consumed by /code-forge:tdd @test-cases.md

Review generated specifications for quality, completeness, and internal consistency:

/spec-forge:review user-login          # Review all specs for user-login

• Checks completeness, internal consistency, specificity, traceability, and actionability
• Compares feature specs against tech-design for API signature and component boundary alignment
• Auto-fixes Critical and Major issues (up to 2 review-fix iterations)
• Leaves <!-- REVIEW: ... --> comments when domain knowledge is needed for a fix
• Automatically runs as Stage 4 in the full chain (/spec-forge <name>)

Best for: After generating specs, before starting implementation.

Audit existing project documentation for quality, completeness, and code alignment:

/spec-forge:audit                          # Audit current project's docs
/spec-forge:audit ../../other-project      # Audit another project

• Cross-references docs against the actual codebase (API surfaces, features, architecture)
• Checks internal consistency between documents (terminology, facts, versions)
• Evaluates quality dimensions (completeness, accuracy, clarity, currency)
• Generates a findings report with severity levels (Critical/Major/Minor/Info)
• Optionally applies fixes to resolved findings

Best for: Single projects with both documentation and source code.

Analyze a collection of documents to understand the knowledge landscape:

/spec-forge:analyze ../../aipartnerup-docs  # Analyze a docs-only repo
/spec-forge:analyze                         # Analyze current project's docs/

• Builds a document map with type classification and theme clustering
• Detects conflicts and contradictions between documents
• Identifies coverage gaps and missing documentation
• Finds redundancies and near-duplicate content
• Assesses document staleness from content signals
• Proposes reorganization when structure can be improved

Best for: Document ecosystems, cross-repo docs, research collections, mixed-format doc repos.

/spec-forge:idea cool-feature              # Brainstorm (iterative, multi-session)
    ↓ (graduated)
/spec-forge cool-feature                   # Idea → Decompose → Tech Design + Feature Specs → Review
    ↓
/code-forge:plan @docs/features/<component-name>.md   # Break into tasks and execute

Quick path (skip idea stage):

/spec-forge:tech-design cool-feature       # Tech Design + auto-generated feature specs
/spec-forge:review cool-feature            # Review specs before implementation
/code-forge:plan @docs/features/<component-name>.md   # Generate implementation plan

Default auto chain (idea → decompose → tech-design):

Idea Draft ──────────────────────────────→ Tech Design ──→ Feature Specs
(ideas/<name>/draft.md)                   §3.5/§3.6/§3.7    docs/features/
                                           populated from      auto-generated
                                           idea draft          in Step 7

With optional upstream docs (on-demand PRD/SRS add traceability):

PRD (optional) ──→ SRS (optional) ──→ Tech Design ──→ Feature Specs
/spec-forge:prd     /spec-forge:srs     §3.5/§3.6/§3.7    docs/features/
                                         traced to           auto-generated
                                         FR/NFR IDs          in Step 7

Each feature gets its own directory under docs/:

• docs/<feature-name>/tech-design.md (always generated)

Auto-generated feature specs go to docs/features/:

• docs/features/overview.md (feature index + dependency graph)
• docs/features/<component-name>.md (per-component implementation spec)

On-demand documents (when explicitly requested):

• docs/<feature-name>/prd.md
• docs/<feature-name>/srs.md
• docs/<feature-name>/test-cases.md

For decomposed projects, a manifest is also generated:

• docs/project-<project-name>.md

Brainstorming ideas are stored in the project's ideas/ directory. Add ideas/ to .gitignore to keep them private, or commit for team collaboration.

code-forge — spec-forge handles upstream specification (what to build and why), code-forge handles downstream execution (how to build it and ship it).

spec-forge works perfectly standalone — code-forge is optional.

If code-forge is not installed, each command's "Next Steps" section provides general guidance for moving forward with implementation.

/plugin install tercel/spec-forge

See .codex/INSTALL.md

See .opencode/INSTALL.md

MIT License