Agentic Coding Playbook

A practical playbook for teams: when to use AI agents, when humans should stay in the loop, and how to make agentic coding safer and more accountable.

Lan Nguyen Si | CC BY 4.0

Browse it with navigation and search on the agent-dx playbook site, linked from the repository README.

Core thesis

Agents may optimize. Humans must remain accountable. Systems must enforce the difference.

This package focuses on the human and organizational side of agentic coding:

• when to use agents
• when not to use agents
• what kinds of review are needed
• how to reduce operational and quality risk

Main playbooks

PLAYBOOK-EN.md: Executive playbook (English)

PLAYBOOK.md: Executive playbook (Deutsch)

One-page overview

English:

Deutsch:

Additional reference material

For teams that want a clearer model for review depth and implementation quality:

• Review Levels and Implementation Standards (English)
• Review-Stufen und Qualitätsstandards (Deutsch)
• Implementation Agent Standard (English)
• Implementierungs-Agent-Standard (Deutsch)

These companion documents describe the difference between no review, normal review, and rigorous review, and propose a practical Role + Skill + Standard model for implementation agents.

For teams remediating dependency CVEs in an npm lockfile specifically:

• npm Lockfile CVE Remediation (English)
• npm Lockfile CVE Remediation (Deutsch)

Operational artifacts

For teams that want to apply this model directly in a repository, this package also contains reusable artifacts:

• templates/: specs with evals, task slicing, agent tasks, risk classification, review reports, and PR descriptions
• policies/: example risk tiers, protected files, approval rules, and quality gates
• skills/: reusable agent role prompts (task slicer, implementation agent, reviewer, skeptic, release gate)
• examples/: placeholder for stack-specific and workflow-specific examples; none are included yet

These artifacts are tool-agnostic and bilingual. Adapt them to your harness, CI system, and team workflow; they are not an enforcement layer on their own.

Related projects

• agent-engineering-playbook: Full technical reference with checklists and templates
• project-forge: Greenfield scaffolding with built-in gates
• depsight: Security and dependency health for brownfield

Contributing

This is a living document. Feedback, corrections and additions are welcome via Issues and Pull Requests.

License: CC BY 4.0