Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/isaackolamide/sdd-harness itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 27, 2026sdd-harness
SDD workflow orchestrator — thin layer that composes
agent-skillsandsuperpowersinto end-to-end development workflows.
What This Is
sdd-harness is a plugin that provides the SDD (Spec-Driven Development) workflow — a structured path from blank slate to shipped feature, with a built-in iteration loop for post-implementation findings:
constitution → task → prd → spec → plan → implement → verify-feature
↓
findings → spec → plan → implement
It wraps skills from two major plugins -> superpowers and agent-skills
Plugin Stack
sdd-harness (orchestrator) — 10 SDD workflow skills
↓ delegates to
agent-skills (primitives) — 24 engineering skills
superpowers (discipline) — TDD, subagent-driven execution, brainstorming
claude-md-management (tooling) — CLAUDE.md audit and improvement
Skills
| Skill | What It Does |
|---|---|
/using-sdd-harness | Routing tree — which skill for which task, across all plugins |
/sdd-constitution | Create or extract SDD constitution — works for new and existing projects |
/sdd-task | Scope a roadmap milestone into one or more right-sized task files — size gatekeeper that splits oversized work before it reaches /sdd-prd |
/sdd-prd | Generate Product Requirements Document (PRD) from a task file or rough idea via interactive discovery |
/sdd-write-spec | Create feature spec for a new feature — links it on the roadmap milestone and, if you want one, drafts a flow diagram (Mermaid or Excalidraw) |
/sdd-plan-feature | Plan a feature from a feature spec file — outputs plan.md/requirements.md/validation.md; triggers ADR for significant arch decisions |
/sdd-implement-plan | Execute a feature plan — 3-way mode (subagent-driven / autonomous / checkpoint), domain-aware dispatch, TDD enforced with committed RED/GREEN evidence, phase checkpoints, developer whole-branch review |
/sdd-implement-parallel-plans | Execute multiple independent feature plans concurrently using isolated git worktrees |
/sdd-verify-feature | Run parallel verification gate (test engineer & code quality reviewer), check TDD evidence, update progress files, run pre-merge audits, and integrate the branch |
/optimise-claude-md | Audit and improve any project's CLAUDE.md |
Installation
SSH blocked? If your network blocks SSH connections to GitHub, run this once before installing:
git config --global url."https://github.com/".insteadOf "git@github.com:"This rewrites any
git@github.com:URLs to HTTPS automatically.
Platform Integration Guides
1. Claude Code
# Register marketplaces
claude plugin marketplace add https://github.com/addyosmani/agent-skills.git
claude plugin marketplace add https://github.com/isaackolamide/sdd-harness.git
# Install plugins
claude plugin install superpowers@claude-plugins-official
claude plugin install agent-skills@addy-agent-skills
claude plugin install claude-md-management@claude-plugins-official
claude plugin install sdd-harness@sdd-harness
2. Antigravity CLI / IDE
agy plugin install https://github.com/addyosmani/agent-skills.git
agy plugin install https://github.com/obra/superpowers.git
agy plugin install https://github.com/isaackolamide/sdd-harness.git
3. GitHub Copilot
copilot plugin install addyosmani/agent-skills
copilot plugin install obra/superpowers
copilot plugin install isaackolamide/sdd-harness
How to Use
The SDD (Spec-Driven Development) workflow is structured into four main phases, executed in sequence.
flowchart TD
Start(["Start Task"]) --> Constitution{"Constitution exists?"}
Constitution -- "No" --> Init["0. Init Constitution: sdd-constitution"]
Init --> Task["0.25. Scope Task: sdd-task"]
Constitution -- "Yes" --> Task
Task -- "One task file per capability" --> PRD["0.5. PRD: sdd-prd"]
PRD --> Spec["1. Spec: sdd-write-spec"]
Spec --> Plan["2. Plan: sdd-plan-feature"]
Plan --> Implement["3. Implement: sdd-implement-plan"]
Implement --> Verify["4. Verify: sdd-verify-feature"]
Verify --> Finish(["Feature Integrated / Done"])
%% Loop for issues found during verification
Verify -. "Verification Failures / Code Quality Issues" .-> Implement
%% Loop for post-integration feedback/bugs
Finish -. "Post-Integration Findings / Bugs" .-> Spec
Finish -. "Next Milestone / New Idea" .-> Task
[!TIP] Unsure which skill to run for a specific task? Run
/using-sdd-harnessat the start of your session to view the authoritative routing tree across all plugins in the stack.
Choose Your Entry Point
Pick the starting command that matches your current project state:
1. Starting Fresh (New Project or Greenfield Initiative)
You do not have a project constitution yet. You need to bootstrap the core scope, guidelines, roadmap, or crystallize a rough idea.
/sdd-constitution # Interactive interview → generates mission.md, tech-stack.md, roadmap.md
/sdd-task # Scope a milestone → sdd-specs/tasks/YYYY-MM-DD-{task}-task.md (splits oversized work)
/sdd-prd # Discovery interview (seeded by task file) → sdd-specs/prds/YYYY-MM-DD-{feature}-prd.md
/sdd-write-spec # Propose new feature spec (reads PRD) → features/YYYY-MM-DD-{feature}-spec.md
/sdd-plan-feature # Choose feature/milestone → plan.md, requirements.md, validation.md
/sdd-implement-plan # TDD slice-by-slice implementation loop
/sdd-implement-parallel-plans # Implement multiple independent plans concurrently
/sdd-verify-feature # Formal validation, quality audits, ticks roadmap, merges branch
Note: For an existing codebase, /sdd-constitution will automatically read your folder structure and commit history to pre-fill context before asking any questions.
2. Constitution Exists (Adding a New Feature)
The project constitution already exists. You are starting a new feature from the roadmap.
/sdd-task # Scope the roadmap milestone → one or more task files
/sdd-prd # Discovery interview per task file → generates PRD
/sdd-write-spec # Creates feature spec (consumes PRD) → sdd-specs/features/YYYY-MM-DD-{feature}-spec.md
/sdd-plan-feature # Reads feature spec → plan.md, requirements.md, validation.md
/sdd-implement-plan # Runs implementation slices & developer review
/sdd-implement-parallel-plans # Implements multiple independent feature plans concurrently
/sdd-verify-feature # Formally validates criteria & integrates branch
Note: If /sdd-task splits a milestone into several task files, run /sdd-prd on them one at a time.
3. Post-Implementation Findings (Bugs & Feedback)
Manual testing or review revealed issues or adjustments after running /sdd-implement-plan. Feed findings back into the spec/plan loop to handle them with discipline (no task file or PRD needed):
/sdd-write-spec # Generate spec from findings seed (inline or --file path/to/notes.md)
/sdd-plan-feature # Plan the fixes → plan.md, requirements.md, validation.md
/sdd-implement-plan # Implement fixes with TDD
/sdd-verify-feature # Validate fixes and complete integration
Command Deep Dive & Outputs
Phase 1: Discovery & Specification (/sdd-constitution, /sdd-task, /sdd-prd, /sdd-write-spec)
Sets product requirements, project-wide boundaries, and creates feature specifications.
/sdd-constitution(No existing specs):sdd-specs/mission.md— Core objective, user persona, and "never do" list boundaries.sdd-specs/tech-stack.md— Directory structure, code style rules (with code snippet), and test runner configurations.sdd-specs/roadmap.md— Project milestones and release phases.
/sdd-task(Roadmap milestone or rough idea):- Runs a size check first. If the milestone bundles more than one actor, workflow, or independently shippable capability, it splits it into one task file per capability before writing anything.
sdd-specs/tasks/YYYY-MM-DD-{name}-task.md— Context, Expected Implementation, Out of Scope, Acceptance Criteria (Given/When/Then, at least two scenarios each for Happy Path, Error Path, and Boundary), Open Questions/Assumptions, and a task-specific Definition of Done.sdd-specs/roadmap.md— One milestone bullet per task file, linking to it.
/sdd-prd(Task file or rough product/feature idea):sdd-specs/prds/YYYY-MM-DD-{name}-prd.md— Comprehensive Product Requirements Document (PRD) detailing user journeys, Acceptance Criteria, MoSCoW priorities, and constraints. When seeded by a task file, it records the path underTask Referenceand still runs the full discovery interview.
/sdd-write-spec(PRD or idea exists):sdd-specs/features/YYYY-MM-DD-{name}-spec.md— Scoped feature specification (can translate a PRD into a technical spec).sdd-specs/diagrams/YYYY-MM-DD-{name}-flow.md— Optional flow diagram. You are asked whether you want one, then pick the detail level (High-level / Standard / Detailed) and the format (Mermaid, or Excalidraw with a companion.excalidrawfile).sdd-specs/roadmap.md— Appends the spec link to the existing task milestone bullet, or adds a new bullet when there is no upstream task file (e.g. bug-fix path).
Phase 2: /sdd-plan-feature
Breaks the feature spec into structured, implementable tasks.
- Outputs:
sdd-specs/plans/YYYY-MM-DD-{name}/plan.md— Task breakdown with strict interface contracts, aTDD: requiredorTDD: N/A — <reason>line per task, and phase checkpoints.sdd-specs/plans/YYYY-MM-DD-{name}/requirements.md— Project scope, out-of-scope items, and design constraints (e.g. security, telemetry, migration risk).sdd-specs/plans/YYYY-MM-DD-{name}/validation.md— Acceptance criteria checklist and definition of done.sdd-specs/docs/decisions/ADR-{NNN}.md— Generated automatically if significant architectural choices surface.
Phase 3: /sdd-implement-plan
Executes the plan slice by slice using Test-Driven Development (TDD) on an isolated feature branch.
- Execution Modes (selected on start):
- Subagent-driven (Recommended for $\ge$ 4 slices): Spawns an isolated subagent per task; maintains context cleanliness.
- Autonomous: Executes the entire plan in a single run without pausing.
- Checkpoint: Pauses for user confirmation after completing each slice.
- Branch isolation: A new feature branch is always created at the start of execution — regardless of the current branch state.
- TDD evidence: For every
TDD: requiredtask, the reviewer rejects the slice unless the implementer's report shows a failing test (RED) before the passing one (GREEN). - Outputs:
- Commits corresponding to each task slice.
sdd-specs/plans/YYYY-MM-DD-{name}/tdd-evidence.md— The committed RED/GREEN record for each task (or its N/A reason), added in the same bookkeeping commit that ticks the task.- Phase checkpoints verified, ticked, and committed at phase boundaries.
- Whole-branch code review results.
Phase 4: /sdd-verify-feature
Validates code compliance and merges the feature branch.
- Workflow:
- Runs a Parallel Verification Gate: dispatches the
test-engineersubagent (givenvalidation.md,requirements.md, andtdd-evidence.md) to verify validation criteria and acode-quality-reviewersubagent to audit code quality simultaneously. - Audits raw execution logs and quality reports. Any issues found are added to
plan.mdunder## Validation Fixesand## Code Quality Review Fixesto be resolved via/sdd-implement-plan. - Ticks off progress in
plan.mdandroadmap.mdupon completion. - Runs a pre-merge programmatic verification gate (checks every plan task has a
tdd-evidence.mdentry, checks git cleanliness, runs main build, lint, and test scripts). - Integrates the branch and cleans up using
superpowers:finishing-a-development-branch.
- Runs a Parallel Verification Gate: dispatches the
What's NOT in This Plugin
Skills that used to be copied here now live in agent-skills directly:
- interview-me, idea-refine
- incremental-implementation, api-and-interface-design
- code-review-and-quality, security-and-hardening
- ci-cd-and-automation, observability-and-instrumentation
- documentation-and-adrs, deprecation-and-migration
Install agent-skills@addy-agent-skills to get all of these.