SDLC Plugins
Private Claude Code plugins for the AI Platform Optimization team.
Setup
Adding the marketplace registers this repo as a plugin source. Plugins are then installed individually — install whichever ones you need.
# 1. Add marketplace (one-time)
/plugin marketplace add https://github.com/inference-sim/sdlc-plugins
# 2. Install plugins (pick what you need)
/plugin install research-ideas@sdlc-plugins
/plugin install hypothesis-test@sdlc-plugins
Auto-Updates
To enable automatic plugin updates when new versions are released, go to /plugin > Marketplaces > sdlc-plugins and enable auto-update
Manual Install / Update / Remove
# Install individual plugins
/plugin install research-ideas@sdlc-plugins
/plugin install hypothesis-test@sdlc-plugins
# Update all installed plugins to latest versions
/plugin marketplace update sdlc-plugins
# Remove a plugin
/plugin uninstall research-ideas@sdlc-plugins
/plugin uninstall hypothesis-test@sdlc-plugins
Available skills
research-ideas
Generates research ideas with multi-model AI review (Claude, GPT-4o, Gemini).
/research-ideas
Features:
- Guided workflow - step-by-step configuration for problem, background, judges
- Multi-source context - gather background from repos, papers, and web search
- Parallel processing - background agents fetch context simultaneously
- Configurable judges - choose which AI models review your ideas
- Progress dashboard - visual task tracking throughout the process
Just run the command and let it guide you. The plugin will:
- Help you create or locate a problem statement
- Gather background from multiple sources (repos, papers, web)
- Configure review judges (3 models by default)
- Generate iteratively-reviewed ideas with visual progress tracking
Outputs a single research.md with problem, background, all ideas, reviews, and executive summary.
Creating a Problem Statement
Create a problem.md file in your project directory. Some questions you can consider to help drive the problem statement:
- Briefly describe the background and motivation
- State what is needed to be solved
- Any limitations or requirements
- How will you define success?
Need help articulating your problem? Use the brainstorming superpower:
/superpowers:brainstorm brainstorm problem statements for the /research-ideas skill
This will help you explore and refine your problem statement before generating research ideas.
hypothesis-test
Guided hypothesis-driven experimentation with auto-analysis.
/hypothesis-test
Features:
- 6-screen guided flow - project setup, hypothesis generation, selection, experiment design, testing, commit
- Parallel hypothesis generation - background agents generate all hypotheses simultaneously
- Parallel experiment scaffolding & testing - each hypothesis gets its own background agent
- Batch approval - review all experiment designs at once before any tests run
- Auto-analysis - experiments run, analyze results, and determine verdicts automatically
Just run the command and it will:
- Ask you to pick a project, focus area, and hypothesis count
- Generate hypotheses in parallel (background agents scan the project for untested behaviors)
- Let you select which hypotheses to test and execution mode (parallel/sequential)
- Scaffold experiments, present designs for batch approval
- Run approved experiments and document findings with verdicts (Confirmed/Refuted/Inconclusive/Failed)
- Optionally commit all results to git
Outputs per-hypothesis directories under hypotheses/ with HYPOTHESIS.md, run.sh, analyze.py, and FINDINGS.md, plus a hypotheses/README.md catalog.
Authentication
For private repo access, use SSH keys or set GITHUB_TOKEN:
export GITHUB_TOKEN=$(gh auth token)
Examples
See example output here.
Contributing
See CONTRIBUTING.md for plugin development.
