Human-on-the-Loop (HOTL) operating model for AI-native development
npx claudepluginhub yimwoo/hotl-pluginHuman-on-the-Loop operating model: skills, workflows, and adapters for AI-native development
Claude Code plugins for the Slidev presentation framework
Bundled plugins for actuating and debugging the Chrome browser.
Production-ready workflow orchestration with 75 focused plugins, 182 specialized agents, and 147 skills - optimized for granular installation and minimal token usage
HOTL (Human-on-the-Loop) is an AI coding workflow plugin and skill pack for Codex, Claude Code, and Cline. It adds design, planning, review, and verification guardrails so AI-generated changes do not land without evidence.
Use HOTL when you want a structured AI development workflow: brainstorm before coding, write a plan before implementation, review risky changes, and verify results before claiming success.
Works with Claude Code, Codex, and Cline. Adapter templates are also available for Cursor and GitHub Copilot.
Most AI coding sessions fail in predictable ways: code starts before requirements are clear, plans skip verification, risky changes execute without review, and the agent claims success without evidence.
HOTL prevents all four by enforcing structured workflows for implementation tasks while staying out of the way for code questions, quick fixes, and debugging.
If someone searches for a "HOTL plugin" or a "Human-on-the-Loop AI coding workflow", this repo is the main project: it contains the canonical HOTL skills, workflow templates, and installation docs for Codex, Claude Code, and Cline.
/plugin marketplace add yimwoo/hotl-plugin
/plugin install hotl@hotl-plugin
git clone https://github.com/yimwoo/hotl-plugin /tmp/hotl-plugin
bash /tmp/hotl-plugin/install.sh --codex-plugin
After install, restart Codex, switch the plugin directory to Local Plugins, and click Add to Codex for HOTL.
For native skills install (local dev, older Codex): clone to ~/.codex/hotl + symlink to ~/.agents/skills/hotl.
Plugin install does not automatically remove an older native-skills install. If both are present, Codex may discover duplicate HOTL sources. See docs/README.codex.md for the recommended migration path.
For native skills installs, ~/.codex/hotl is the stable channel and should track origin/main. Restart Codex after install. Full guide: docs/README.codex.md
curl -fsSL https://raw.githubusercontent.com/yimwoo/hotl-plugin/main/install-cline.sh | bash
Full guide: docs/README.cline.md
Implementation tasks follow seven phases:
| Phase | What happens |
|---|---|
| Brainstorm | Clarify requirements. Compare approaches. Define intent, verification, and governance contracts. |
| Plan | Generate a hotl-workflow-<slug>.md with steps, verification, loop conditions, and gates. |
| Lint | Self-check built into planning. Structural lint runs automatically in execution preflight. |
| Branch | Create an isolated git branch. Dirty repos hard-fail. |
| Execute | Run the plan in loop, manual, or subagent mode. |
| Review | Review findings are checked against the codebase and HOTL contracts before acting. |
| Verify | Run tests, lint, and verify commands. No green light without proof. |
Here is what a real HOTL feature-delivery session can look like:
Execution Summary
| Step | Status | Iterations |
|-----------------------------------------------|--------------------|------------|
| Step 1: Add feature flag and config wiring | Done | 1 |
| Step 2: Add backend endpoint for saved views | Done | 2 |
| Step 3: Add database migration and model | Done | 1 |
| Step 4: Build saved views panel UI | Done | 3 |
| Step 5: Connect UI to API state flow | Done | 2 |
| Step 6: Add analytics + audit logging | Done | 1 |
| Step 7: Add unit tests for reducers/hooks | Done (28/28) | 2 |
| Step 8: Add API integration tests | Done (12/12) | 2 |
| Step 9: Add e2e coverage for create/apply | Done (6/6) | 3 |
| Step 10: Run lint and typecheck | Done | 2 |
| Step 11: Run full test suite | Done (46/46) | 1 |
| Step 12: Human review and acceptance | Approved | 1 |
9 files modified, 1 migration added, 3 new test files. Unit, integration, and e2e suites all passing.
Every step has a verify command. If verification fails, execution stops and reports instead of silently claiming success.