Production skills for Claude Code that output artifacts, not advice.
These aren't prompt wrappers. They spawn parallel agents, search the web for real data, and produce structured outputs — social simulations and complete TypeScript CLIs from any codebase.
Built while running multiple SaaS products from the terminal.
npx skills add progrmoiz/skillsOther install methods
Individual skills:
npx skills add progrmoiz/skills/simulateClaude Code plugin system:
/plugin marketplace add progrmoiz/skills
/plugin install progrmoiz-skills
Manual (clone):
git clone https://github.com/progrmoiz/skills.git ~/.claude/skills/progrmoiz-skills| Skill | What it does | Install |
|---|---|---|
| simulate | 12 personas react to your announcement over 3 rounds — form camps, shift opinions, surface objections | npx skills add progrmoiz/skills/simulate |
| cli-generate | Point at any codebase, get a complete TypeScript CLI with --json, auth, doctor command |
npx skills add progrmoiz/skills/cli-generate |
| estimate | AI-era time estimates calibrated with research data from Anthropic, METR, and BSWEN studies | npx skills add progrmoiz/skills/estimate |
Simulate how 12 diverse personas react to your announcement before you ship it. Each persona has a unique profile (MBTI, influence weight, sentiment bias) and does their own web research for grounding. Over 3 rounds, they read each other's reactions, form camps, and shift opinions.
Produces: Verdict, opinion trajectory table, camp analysis, top 3 ranked objections with fixes, suggested rewrite.
When to use: Before pricing changes, product launches, policy shifts, layoffs, or any high-stakes public message.
How it works:
- You paste the announcement
- 12 personas are generated (5 mandatory archetypes + 7 dynamic based on announcement type)
- Round 1 — each reacts independently (12 parallel agents, each searches the web)
- Round 2 — each reads ALL Round 1 reactions, can shift stance, replies form (12 parallel agents)
- Round 3 — final positions harden after seeing the full debate (12 parallel agents)
- Report synthesizes everything into actionable insights
~36 agent calls. The magic is in Round 2 — that's where agents react to each other and camps emerge.
Auto-generate a complete, production-grade TypeScript CLI from any codebase. Works with MCP servers, OpenAPI specs, Next.js, Express, Flask, Go, Rails, gRPC, GraphQL — or any API in any language.
Produces: Working CLI with Commander.js, dual output (pretty for humans, JSON for agents), multi-account auth, doctor command, single esbuild bundle, README, and SKILL.md.
When to use: When you want to make any web service agent-ready, or when you need a CLI for your API.
How it works:
- Detect — scans codebase for endpoints (Tier 1: pattern matching for known frameworks, Tier 2: reads source code for any language)
- Plan — maps endpoints to CLI commands, presents for approval
- Scaffold — generates fixed infrastructure (auth, output, spinner, table renderer)
- Generate — creates one command file per endpoint
- Verify — builds, runs
--help, checks bundle size
5 production deps max. Hand-rolled output. No chalk, no ora, no boxen.
AI-era time estimates calibrated with real research data. Breaks tasks into subtasks with pre-AI vs post-AI effort, accounting for third-party API complexity and research-backed multipliers.
Produces: Subtask breakdown with individual estimates, assumptions, complexity factors.
When to use: Scoping work, giving time estimates to stakeholders, comparing pre-AI vs post-AI effort.
| Prompt-based skills | These skills | |
|---|---|---|
| Agents | Single prompt, single response | 12–36 parallel agents per run |
| Web access | None | Searches Reddit, Twitter, competitor sites for real data |
| Output | Paragraphs of advice | Tables, scorecards, verdicts, working code |
| Phases | One-shot | Multi-phase with synthesis between each stage |
| Interaction | Agent talks to you | Agents talk to each other (simulate Round 2) |
Have a skill that produces artifacts instead of advice? PRs welcome.
MIT