Agents Playbook
v0.1 · CC-BY-4.0

The gold-standard playbook for shipping production software with AI coding agents.

Pillars, patterns, prompts, and gates earned from a year of agent-driven development on a real production codebase. Drop-in templates, ready-to-run gate scripts, and a structure built for both humans and agents.

Read the playbookStart from CLAUDE.mdBundle (.zip)
CLAUDE.md
# CLAUDE.md

## Non-negotiables

1. No `any`. Schema parse every boundary.
2. Named exports only.
3. Typed AppError + stable codes.
4. createLogger(tag); no console.log.
5. ADR before architecture. RFC before
   breaking contracts.
6. Ship complete or don't ship.
7. Merges sum work, never subtract.
8. UI via shared primitives only.
9. Every visible string is intl.
10. Tokens; no raw color literals.

## Before you ship
```
pnpm check:quality-gates
pnpm check:all
```
6
Pillars
70+
Patterns
13
Gate scripts
6
SDLC phases
Why

Built for the kind of code agents actually ship.

Production-earned

Every rule traces to a reproducible failure mode. No theory; only patterns paid for in shipped repos.

Dual-mode docs

Each page has a Human TL;DR and a For-Agents section. Optimised for both linear reading and RAG retrieval.

Gates included

13 reference gate scripts (Node 22, zero deps) you can drop into any repo. Pure copy-paste.

Drop-in templates

ADR, RFC, PR-intent, CLAUDE.md, AGENTS.md, MEMORY.md — ready for any project.

Pillars

Six pillars. Six SDLC phases. One matrix.

Each pillar carries a universal (stack-agnostic) layer plus concrete recipes for a TypeScript stack. Together they form a checklist for any agent-augmented team.

19 docs

Architecture

Modular boundaries, typed contracts, ADRs / RFCs, file-size budgets, anti-overengineering, distributed data, event streaming, multi-region.

Explore
17 docs

Security

RBAC, vault, audit ledger, egress allowlist, vulnerability mgmt, secrets deep, multi-tenant isolation, compliance (SOC 2 / GDPR), AI/LLM safety.

Explore
9 docs

UI / UX

Design tokens, primitives catalog, intl + ICU, empty states, a11y (WCAG-AA deep), motion + reduced-motion, whitelabel, design-system governance.

Explore
14 docs

Quality

Test pyramid, gates, sanity, mutation, observability + SLOs, performance budgets, chaos engineering, CI/CD pipeline, FinOps, contract testing.

Explore
5 docs

Governance

PR intent manifest, merge rules, tombstones, phased PRs — keeps multi-author work additive instead of subtractive.

Explore
6 docs

AI Collaboration

Bootstrap docs (CLAUDE.md / AGENTS.md), persistent memory, sub-agent recipes, slash commands, concurrent-agent survival.

Explore
For agents

Built so your agents can find what they need.

  • 1/llms.txt at root — convention for LLM-readable site map.
  • 2Each doc serves a raw .md alongside the HTML — agents fetch source.
  • 3Structured data (JSON-LD) and OpenGraph on every page.
  • 4Single-file llms-full.txt bundles every doc for one-shot RAG indexing.
  • 5Zip bundle (/playbook-bundle.zip) for download + local indexing.
# fetch raw markdown from any doc
curl https://playbook.agentskit.io/raw/pillars/architecture/universal.md

# one-shot bundle for RAG indexing
curl https://playbook.agentskit.io/llms-full.txt

# zip everything
curl -O https://playbook.agentskit.io/playbook-bundle.zip

Start with the eight non-negotiables.

The kernel of the playbook. If an agent breaks one, fail the PR. Everything else flows from these.

Read the playbookSee the matrix