projects / hkit
hkit
Spec-driven AI Engineering Framework
Design and implementation
What it is
hkit is an additive framework around GitHub Spec Kit. It connects specification, clarification, planning, task breakdown, implementation, review, architecture decisions, verification, and delivery across Claude Code, Codex, and other coding agents. It is not an official GitHub Spec Kit fork and does not imply upstream endorsement.
The problem
Long AI coding sessions can produce changes that are hard to review, test, or hand off. hkit structures the work into planned stages with human gates and evidence-based outputs.
What I worked on
- Built specification, planning, implementation, review, and delivery workflows.
- Added behavioral quality gates and negative controls.
- Created architecture decision record capture.
- Implemented review routing and agent handoffs.
- Added session hooks and worktree isolation.
- Shared instructions across coding agents.
- Built Markdown to HTML/PDF delivery with Mermaid diagrams.
- Supported Arabic RTL output and redacted client deliverables.
Decisions
Additive around Spec Kit
The framework wraps untouched upstream Spec Kit assets rather than forking or modifying them.
Human gates at each stage
Specification, planning, implementation, and delivery each have explicit review points outside the model's judgment.
Evidence-based handoffs
Each stage produces artifacts that the next stage can verify against, not just chat history.
Testing and verification
- Behavioral quality gates and negative controls on agent outputs.
- Review routing before changes are accepted.
- Structured handoff artifacts between workflow stages.
Stack
Workflow
Specification · Clarification · Planning · Task decomposition · Implementation · Review · Verification · Delivery
Agent tooling
Claude Code · OpenAI Codex · MCP · Subagents · Custom skills · Hooks · Git worktrees
Output
Markdown · Mermaid diagrams · HTML/PDF delivery · Arabic RTL
Status
Private personal project. No public repository link at this time.