Portable agent skills by Greg Lichtenthal for people who want AI to strengthen human creativity, judgment, and agency.
This catalog is built around judgment infrastructure for human-led AI work: organize messy context, pressure-test the thinking, deliberate consequential decisions, and test ideas before trusting them.
Read the short thesis: Judgment Infrastructure for Human-Led AI
Start with the guided workflow: Start Here
See the full loop in one scenario: Judgment Infrastructure Demo (Markdown source)
Browse practical use cases: Product Strategy · Founder Decisions · Research Synthesis · Career Decisions
Awesome list: awesome-judgment-infrastructure
The suite installer downloads the pinned stable release of the four core judgment skills into Codex's current user-skills directory:
git clone https://github.com/glichtenthal/agent-skills.git
cd agent-skills
bash install.sh --target codexUse --dry-run to preview the installation, --update to replace existing copies, or --skill the-quorum to install one skill. AI Project Steward installs separately from the four-part judgment suite and is recommended for people who regularly build with AI; install it with bash install.sh --skill ai-project-steward. Run bash install.sh --help for all options.
For ChatGPT or Claude, import each .skill release through the product's Skills settings. The shell installer is for Codex.
For Product Design Director and Production Frontend Engineer alongside Steward, use the separate product development installation guide.
You do not need the full loop every time. Start with the part that matches what you have.
| If you have... | Use... | Output |
|---|---|---|
| Messy notes, transcripts, research, or context | The Briefing Room | A brief you can think with |
| A claim, plan, draft, or idea that needs honest critique | Ground Truth | Weak assumptions, missing evidence, and sharper next steps |
| A consequential decision with real trade-offs | The Quorum | Structured disagreement and a calibrated recommendation |
| An idea, claim, or decision that needs proof before commitment | Test Drive | The smallest credible test and the signal that would change your mind |
Turn messy context into a brief you can think with.
- Install: https://github.com/glichtenthal/briefing-room/releases/download/v1.2.1/briefing-room.skill
- Best for: messy notes, transcripts, research dumps, meeting context, customer feedback, and personal sensemaking.
- Quick demo: https://github.com/glichtenthal/briefing-room/blob/main/examples/quick-demo.md
Calibrated honesty and anti-sycophancy for plans, decisions, reviews, and ideas.
- Install: https://github.com/glichtenthal/ground-truth/releases/download/v1.2/ground-truth.skill
- Best for: pressure-testing ideas, reviewing work, avoiding easy validation.
- Quick demo: https://github.com/glichtenthal/ground-truth/blob/main/examples/quick-demo.md
A five-member expert council that pressure-tests consequential decisions.
- Install: https://github.com/glichtenthal/the-quorum/releases/download/v1.4.1/the-quorum.skill
- Best for: strategy calls, hiring decisions, build-vs-buy choices, and other consequential trade-offs.
- Quick demo: https://github.com/glichtenthal/the-quorum/blob/main/examples/quick-demo.md
Test an idea before you trust it.
- Install: https://github.com/glichtenthal/test-drive/releases/download/v1.6.1/test-drive.skill
- Best for: testing ideas, messages, prompts, skills, strategies, decisions, and data claims before overcommitting.
- Quick demo: https://github.com/glichtenthal/test-drive/blob/main/examples/quick-demo.md
These skills are designed for Claude and Codex-style agent workflows. Each repository includes its own installation instructions and Codex interface metadata.
- The Briefing Room organizes messy context into a structured brief.
- Ground Truth pressure-tests the brief, claim, plan, or draft.
- The Quorum convenes multiple expert lenses for consequential decisions.
- Test Drive turns ideas, claims, and decisions into small evidence-seeking trials.
Ready to build? AI Project Steward helps the human and AI manage the build as a whole, from visible choices and approvals through decision continuity and verification.
Applied Judgment Systems carry human-led judgment into a specific domain or stage of work.
A clear operating layer for visibility, guidance, approval, continuity, and verification throughout an AI-assisted build.
Use it independently or as the human-led governance and verification layer within a broader AI build harness. It complements the model, tools, runtime, project context, and platform mechanics rather than replacing them.
- Install: https://github.com/glichtenthal/ai-project-steward/releases/latest
- Best for: human-led AI builds where data, cost, security, portability, deployment, or maintenance choices matter.
- Four scenarios: https://github.com/glichtenthal/ai-project-steward/tree/main/examples
- Evaluation evidence: https://github.com/glichtenthal/ai-project-steward/tree/main/evals
Two complementary skills work alongside AI Project Steward:
- Product Design Director defines the product experience, workflows, interaction design, and visual direction, then critiques the rendered result.
- Production Frontend Engineer implements the interface and checks its behavior, visual fidelity, and technical readiness.
Install all three for the combined workflow; use the roles each task needs. A narrow frontend fix does not require a new design cycle.
Install the product development skills
The three product development skills are MIT-licensed, with copyright notices naming Greg Lichtenthal:
MIT permits personal and commercial use, modification, and redistribution, including within an employer's teams. Copies or substantial portions must retain the applicable copyright and license notices. MIT does not require prominent creator credit in a product or marketing, and does not require private modifications to be published.
For other skills linked from this catalog, consult the license in the corresponding repository or release. These skill licenses do not establish a blanket license for every catalog page, asset, or external resource.