Agent skills for turning proven work processes into executable AI-agent workflows.
This is a practical library for people who want agents to do structured work instead of producing loose advice. It includes workflows for productivity, goal execution, process improvement, product validation, design, tactical decisions, research-paper writing, conversation organization, and skill-library maintenance.
| Skill | Use it for |
|---|---|
gtd-workflow |
turn messy tasks, notes, and obligations into trusted next actions |
okr-execution |
convert goals into objectives, measurable key results, initiatives, and check-ins |
pdsa-improvement-loop |
run small continuous-improvement experiments |
dmaic-process-improvement |
improve existing processes with measurable defects or variation |
design-thinking-sprint |
solve user-centered problems through research, prototypes, and tests |
lean-startup-validation |
test startup/product assumptions before scaling |
ooda-decision-loop |
make fast decisions under uncertainty and update from feedback |
research-paper-pipeline |
write a source-grounded paper from legal literature search to final manuscript |
paper-to-ppt-pipeline |
turn a research paper into a structured, styled PPT deck |
conversation-declutter |
turn long or messy conversations into reusable summaries, maps, and resume context |
conversation-to-skill-miner |
scan conversation archives and topic memory for safe reusable skill candidates |
github-skill-adoption-audit |
audit stars, forks, clones, views, and discovery signals for public skill repos |
codex-reliability-troubleshooter |
diagnose Codex context, instruction, approval, performance, and scope friction |
Every skill is supervised by the internal 马斯克 engineering operator pattern:
- question requirements
- delete unnecessary work
- simplify before optimizing
- accelerate feedback loops
- automate last
That means the agent should not blindly run a framework. It should first ask whether the work should exist at all.
Skills that mine private conversations should generate local candidate reports first. Do not publish raw chats, private profiles, credentials, or sensitive examples. Public uploads should contain only generalized workflow instructions, sanitized references, and reusable scripts.
User:
My task list is a mess. Turn it into something I can actually execute.
Agent using $gtd-workflow:
Deleted:
- "Research everything about marketing" because it is too vague.
Project:
- Launch first public post about agent skills.
Next actions:
- Draft one 600-word post.
- Pick 3 GitHub repo links to include.
- Publish to one channel today.
Waiting for:
- GitHub profile README update.
More examples are in examples/.
mkdir -p ~/.codex/skills
ln -s "$PWD/gtd-workflow" ~/.codex/skills/gtd-workflow
ln -s "$PWD/okr-execution" ~/.codex/skills/okr-execution
ln -s "$PWD/pdsa-improvement-loop" ~/.codex/skills/pdsa-improvement-loop
ln -s "$PWD/dmaic-process-improvement" ~/.codex/skills/dmaic-process-improvement
ln -s "$PWD/design-thinking-sprint" ~/.codex/skills/design-thinking-sprint
ln -s "$PWD/lean-startup-validation" ~/.codex/skills/lean-startup-validation
ln -s "$PWD/ooda-decision-loop" ~/.codex/skills/ooda-decision-loop
ln -s "$PWD/research-paper-pipeline" ~/.codex/skills/research-paper-pipeline
ln -s "$PWD/paper-to-ppt-pipeline" ~/.codex/skills/paper-to-ppt-pipeline
ln -s "$PWD/conversation-declutter" ~/.codex/skills/conversation-declutter
ln -s "$PWD/conversation-to-skill-miner" ~/.codex/skills/conversation-to-skill-miner
ln -s "$PWD/github-skill-adoption-audit" ~/.codex/skills/github-skill-adoption-audit
ln -s "$PWD/codex-reliability-troubleshooter" ~/.codex/skills/codex-reliability-troubleshooterRestart Codex after installation.
- Engineering operator /
马斯克: https://github.com/walterkken/engineering-operator-agents - MATLAB and Simulink automation: https://github.com/walterkken/matlab-simulink-agent-skill
- Nobel process skills: https://github.com/walterkken/nobel-process-skills
- Engineering software skills: https://github.com/walterkken/doc-driven-engineering-software-skills
This repository is published as an ownership and development record. No open-source license is granted unless a separate license file is added for a specific skill or release.