A living resource for MLOps practitioners using coding agents in production. This repository combines techniques and guides from the Coding Agents Conference (March 3, 2025) and the ongoing Coding Agents Lunch & Learn series, hosted by the MLOps Community.
"I've installed these skills, no idea if they are making my agent's better. But they're not making them worse!" - @missberg
Not theory — this is what practitioners actually do, day-to-day, with Claude Code and other agents.
16 production-ready skills extracted from conference talks and lunch & learn sessions, plus deep-dive guides, reproducible workflows, and quick-reference resources.
This is a living knowledge base: new session summaries, workflows, and insights are added regularly as the Lunch & Learn series continues.
- Watch the Conference (Full Livestream) — March 3, 2025 at Computer History Museum
- View Conference Agenda
- Full Conference Transcript — 6h 45m, 1159 paragraphs
All 16 installable Claude Code skills, organized by source:
| Skill | What It Teaches | Source | Key Speaker(s) |
|---|---|---|---|
| Visual Regression | Screenshot walkthroughs during dev + AI-powered visual diff at release | Conference | Rob |
| CRISPI Planning | Multi-stage planning pipeline with fresh context windows per stage | Conference | Dex (HumanLayer), Sid (Anthropic) |
| Context Window Management | Instruction budgets, utilization thresholds, saving state to files | Conference | Dex (HumanLayer), Harrison Chase (LangChain) |
| Adversarial Code Review | Three-agent review (builder, critic, meta-reviewer) with confidence filtering | Conference | Sid (Anthropic), Ankit (Databricks) |
| Hooks & Enforcement | Using Claude Code hooks to enforce what CLAUDE.md alone cannot | Conference | Sid (Anthropic), Josh, Milan (Semgrep) |
| Parallel Agent Management | Worktrees, containers, sub-agent contracts for running multiple agents | Conference | Sid (Anthropic), Rob, Harrison Chase (LangChain) |
| Objective Research | Separating intent from investigation to get unbiased codebase research | Conference | Dex (HumanLayer) |
| Expert Persona Skills | Creating short skills that activate latent domain knowledge | Conference | Rafael, Sid (Anthropic) |
| Voice-First Planning | Using speech-to-text for richer initial specs and ideation | Conference | Josh |
| Agent-Maintained Docs | File headers + folder READMEs enforced by hooks for agent navigation | Conference | Rob |
| Three-Layer Memory | MCP memory + auto-memory + external editor as tiered knowledge system | Conference | Audience member |
| Product Research | Two-phase research: cheap tools for gathering, Claude for synthesis | Conference | Rob, Omeh |
| Brainstorming Planner | Structured brainstorming with agent disagreement and decision frameworks | Lunch & Learn | Demetrios |
| Merge Conflict Resolution | Systematic approach to resolving multi-branch conflicts at scale | Lunch & Learn | Rahul |
| Documentation-First Setup | Building tight feedback loops between docs and agent behavior | Lunch & Learn | Leo |
| Validation Runner | Composable validation pipelines with agent-friendly error reporting | Lunch & Learn | Demetrios |
Beyond the 16 skills, this repo contains practitioner guides, reproducible workflows, and quick-reference resources. These are designed to be read (not installed) and provide context for how to apply the skills.
Longer-form resources for understanding broader concepts:
- 10 Rules for Practitioners — Principles learned from production deployments
- Prompting Patterns — Techniques for crafting agent instructions
- Team Scaling Playbook — How to run multiple agents across teams
Reproducible workflows you can adapt for your own projects:
- Plan → Build → Punch Pipeline — Three-stage workflow for structured delivery
- Opus Orchestrator Pattern — Multi-model orchestration strategy
- Parallel Agents with Broomy — Running agents in parallel safely
- Playground Debugging Flow — How to debug agent behavior iteratively
- Validation Sub-Agent Pipeline — Composable validation architecture
- Automated Code Review — End-to-end code review pipeline
Quick-reference resources for when you need answers fast:
- Tools & Plugins Directory — Catalog of available tools and integrations
- Pain Points & Solutions — Common problems and how practitioners solved them
- Top Questions & Answers — FAQ from the community
Pick the skills you want and copy them to your personal skills directory:
# Copy a single skill
cp -r skills/adversarial-code-review ~/.claude/skills/
# Copy several
cp -r skills/crispi-planning skills/context-window-management skills/hooks-and-enforcement ~/.claude/skills/cp -r skills/* ~/.claude/skills/git clone https://github.com/mlopscommunity/Coding-Agents-Conference-skills.git ~/projects/agents-knowledge-base
ln -s ~/projects/agents-knowledge-base/skills/* ~/.claude/skills/Once installed, skills are available in Claude Code:
- Invoke directly:
/skill-name(e.g.,/adversarial-code-review) - Let Claude find them: Claude searches installed skills when relevant to your task
- Reference in CLAUDE.md: Point your project's CLAUDE.md to specific skills
The guides, workflows, and references don't get installed — they live in the repo and you read them as needed. You can bookmark or clone this repo to have them always handy.
For solo developers:
- Read Plan → Build → Punch Pipeline to understand the workflow
- Install Context Window Management and Hooks & Enforcement — prevents the most common mistakes
- Install CRISPI Planning — structured approach to any non-trivial task
- Read 10 Rules for Practitioners to calibrate expectations
For teams:
- Start with solo path above
- Install Adversarial Code Review — automated PR review with noise reduction
- Install Parallel Agent Management — multiply throughput with worktrees
- Read Team Scaling Playbook for team-specific patterns
- Use Opus Orchestrator Pattern for multi-model coordination
Coding Agents: The AI Driven Developer Conference was held on March 3, 2025 at the Computer History Museum in Mountain View, CA. The full-day event brought together builders, founders, and ML practitioners working on coding agents in production.
| Speaker | Company / Role | Talk |
|---|---|---|
| Demetrios Brinkmann | MLOps Community (Host) | Opening — The Era of Side Quests |
| Sid Bidasaria | Co-creator, Claude Code (Anthropic) | How I Use Claude Code — Fireside Chat |
| Scott Breitenother | Co-founder, Kilo Code | Scaling AI-Assisted Development — Lessons from 25 Trillion Tokens |
| Harrison Chase | Founder, LangChain | The Agent Harness — Building General Purpose Agents |
| Sam Partee | CTO, Arcade.dev | General Purpose Agents — Auth, Identity & Tool Use |
| Dexter Horthy | Founder, HumanLayer | The CRISPI Planning Pipeline — Structured Agent Workflows |
| Jess Wang | BrainTrust | Evals for AI Tooling — Quantifying Ship Decisions |
| Ankit Mathur | Databricks | Mosaic Coding Agent — Enterprise AI Development at Scale |
| Zach Lloyd | Founder & CEO, Warp | Agent Orchestration — The Year of Multi-Agent Workflows |
| Speaker | Company / Role | Talk |
|---|---|---|
| Faye Zhang | Staff AI Engineer & Tech Lead, Pinterest | Productionizing Sub-Agents for LLM Post-Training |
| Yannis He | Co-founder, SWE-bench Pro (Scale AI) | The Next Generation of Coding Agent Benchmarks |
| Erin Ahmed | Head of Product, Cleric | Building a Learning Agent — Stateful AI SRE |
| Milan Williams | Senior Product Manager, Semgrep | Three Security Tips for Coding Agents |
| Ash Lewis | CEO, Fastino Labs | Choosing & Maintaining the Right Model for Agents |
| Workshop | Instructor |
|---|---|
| 5 Prompts to Ship Production Code 2x Faster With AI | Mihail Eric |
| Optimizing Codebases for Agents | Shrivu Shankar |
| Software Libraries with No Code | Drew Breunig |
The afternoon featured open unconference sessions where attendees shared techniques live. Contributors whose insights made it into these skills include: Rob, Josh, Rafael, Chad, Yari, Omeh, Niels, and others.
An ongoing series of 90-minute deep dives on specific agent patterns and techniques. Hosted by Demetrios Brinkmann, Rahul Parundekar, and Leo Walker from the MLOps Community.
Session transcripts and summaries are available in docs/lunch-and-learn/. New sessions are added regularly — this is a living resource.
Current sessions included in this knowledge base:
- Brainstorming with agents
- Merge conflict resolution at scale
- Documentation-first development patterns
- Validation and testing pipelines
Each skill's SKILL.md file includes an Attribution section that credits the specific speaker(s), session, and timestamps where the technique was shared.
The full conference transcript is available at docs/coding_agents_conference_transcript.md (6h 45m, 1159 paragraphs).
AI Agents Summit — Seattle — April 17th We're organizing the next agent-focused conference. Early bird tickets available at https://luma.com/ai-agents-summit-seattle
Keep an eye on mlops.community/events for upcoming Lunch & Learn sessions.
The MLOps Community is a global network of ML practitioners, engineers, and data scientists. We focus on the practical side of machine learning operations, AI systems, and agent-driven development.
- Website: https://mlops.community
- Community: https://community.mlops.community (Slack, discussions, events)
- Events: Conferences, lunch & learns, and virtual meetups year-round
This knowledge base is maintained by the community. Contributions, corrections, and new insights are welcome.