TechKnowmad AI Skills

20 enterprise-grade skills for autonomous AI/ML research lab operations. Custom-built skills + composable combination skills. Designed for Claude Code.

What's Inside

Custom Skills (10) — Purpose-Built

Original skills addressing gaps in the Claude Code ecosystem:

Skill	What It Does	Priority
research-commander	Full research lifecycle orchestrator — literature review → hypothesis → experiment → paper → submission	P0
swarm-commander	Multi-agent swarm coordination with consensus protocols and result aggregation	P0
model-evaluator	Automated benchmark suite runner with leaderboards and regression detection	P0
agent-debugger	Trace agent decision chains, replay failures, diagnose root causes	P1
self-healing-agent	Autonomous recovery — detect degradation, auto-fix, escalate when needed	P1
prompt-evolution	Genetic algorithm-based prompt optimization across generations	P1
skill-factory	Meta-skill that auto-generates new skills from detected patterns	P1
patent-researcher	AI/ML patent prior art search, claim drafting, FTO analysis	P2
cost-optimizer	Cross-cloud spend analysis, GPU strategies, token cost reduction	P2
knowledge-distiller	Compress large documents into optimal skill prompts	P2

Combination Skills (10) — Composable Pipelines

Skills composed from multiple base skills for compound capabilities:

Skill	Composes	What It Does
multi-agent-research-swarm	Research Commander + Swarm Commander + arXiv	10 agents review papers in parallel, synthesize into one review
red-team-ai-models	Pen Tester + AI Security + Model Evaluator	Adversarial security testing for your own ML models
self-evolving-skill-ecosystem	Skill Factory + Workflow Miner + Telemetry	Skills that create skills that create skills
academic-output-pipeline	Research Commander + Paper Writer + Grant Writer	Experiment results → paper → poster → slides → grant proposal
antifragile-ai-ops	Chaos Engineer + Self-Healing + Debugger	Break agents, watch them self-recover, learn from it
pareto-prompt-optimizer	Prompt Evolution + Model Evaluator + Cost Optimizer	Find Pareto-optimal prompts (accuracy vs cost vs latency)
zero-touch-ml-pipeline	Data Pipeline + n8n + Ruflo + Self-Healing	Fully autonomous data → train → deploy → monitor → retrain
startup-creative-automation	Brand Architect + fal.ai + Presentation Architect	Brand identity → visual assets → pitch deck in one pipeline
compliance-as-code	Compliance Automator + GRC + Audit Support	Continuous SOC2/ISO/GDPR compliance with auto-evidence
ip-aware-competitive-intel	Patent Researcher + Literature + Competitive War Room	Patents + papers + products in one competitive landscape view

Installation

As a Claude Plugin

claude plugin add TECHKNOWMAD-LABS/techknowmad-ai-skills

Quick Install (All Skills)

chmod +x scripts/install-techknowmad-skills.sh
./scripts/install-techknowmad-skills.sh

Manual Install (Individual Skills)

# Copy a single skill
cp -r custom-skills/research-commander ~/.claude/skills/research-commander

# Or symlink (preserves git updates)
ln -s "$(pwd)/custom-skills/research-commander" ~/.claude/skills/research-commander

Restart Claude Code

# Close and reopen Claude Code to load new skills
claude

Prerequisites

These skills are designed to work with the TechKnowmad MCP stack. They reference MCPs like arXiv, fal.ai, Langfuse, Ruflo, etc. Skills will still function without all MCPs installed — they gracefully fall back when a referenced MCP isn't available.

Repo Structure

techknowmad-ai-skills/
├── .claude-plugin/          ← Plugin metadata
│   ├── plugin.json
│   └── marketplace.json
├── custom-skills/           ← 10 original skills
│   ├── research-commander/SKILL.md
│   ├── swarm-commander/SKILL.md
│   ├── model-evaluator/SKILL.md
│   ├── agent-debugger/SKILL.md
│   ├── self-healing-agent/SKILL.md
│   ├── prompt-evolution/SKILL.md
│   ├── skill-factory/SKILL.md
│   ├── patent-researcher/SKILL.md
│   ├── cost-optimizer/SKILL.md
│   └── knowledge-distiller/SKILL.md
├── combination-skills/      ← 10 composed skills
│   ├── multi-agent-research-swarm/SKILL.md
│   ├── red-team-ai-models/SKILL.md
│   ├── self-evolving-skill-ecosystem/SKILL.md
│   ├── academic-output-pipeline/SKILL.md
│   ├── antifragile-ai-ops/SKILL.md
│   ├── pareto-prompt-optimizer/SKILL.md
│   ├── zero-touch-ml-pipeline/SKILL.md
│   ├── startup-creative-automation/SKILL.md
│   ├── compliance-as-code/SKILL.md
│   └── ip-aware-competitive-intel/SKILL.md
├── scripts/
│   └── install-techknowmad-skills.sh
├── manifest.json            ← Skill registry
├── CONTRIBUTING.md
├── SECURITY.md
└── LICENSE

Why This Exists

For teams: A curated, version-controlled skill ecosystem — shareable across team instances, iteratively improvable.

For the community: Reference architecture for building a research lab skill stack. These skills represent patterns that don't exist elsewhere:

• Multi-agent research swarms (parallel paper review with consensus synthesis)
• Self-evolving skill ecosystems (recursive self-improvement)
• Antifragile AI ops (chaos engineering for agents)
• Pareto prompt optimization (multi-objective prompt engineering)
• Zero-touch ML pipelines (fully autonomous model lifecycle)

Security

See SECURITY.md for vulnerability disclosure policy.

License

MIT — Use, modify, distribute freely. Attribution appreciated.

Contributing

PRs welcome. Each skill must include:

• SKILL.md with Role, Trigger Conditions, and structured instructions
• Integration points listing required MCPs
• At least one example invocation

---

Built by TechKnowmad AI — Powering autonomous intelligence development.