Agents Playbook turns hard-won engineering lessons into repeatable coding-agent behavior. Use its open rules, prompts, memory, evals, and executable gates to establish project standards, design package boundaries, review agent-authored changes, and enforce quality before merge—without fine-tuning model weights or inventing a governance system from scratch.
It is intended for engineering teams adopting coding agents, maintainers standardizing many repositories, and agents that need structured, retrievable operational context.
What is verified
The repository generates and checks its own claims from source:
| Surface | Current proof |
|---|
| Production patterns | 89 |
| Engineering pillars | 6 |
| SDLC phases | 6 |
| Copy-ready templates | 6 |
| Zero-dependency gate scripts | 13 |
| Human and agent guides | 134 |
| Deterministic local answers | 156 |
| Doc Bridge health | 100/100 · A |
The source-of-truth counts live in app/stats.snapshot.json, the local answer catalog in public/deterministic/knowledge.json, and documentation ownership in doc-bridge.config.json. CI rejects drift.
Start in five minutes
Requirements: Node.js 22 and pnpm 10 (the exact version is pinned in packageManager).
pnpm install --frozen-lockfile
pnpm dev
Open http://localhost:3000. If you are new to agent governance, follow Getting started, copy the AGENTS.md template, and add one relevant gate before expanding the policy set.
Run the fast gates on every commit with the versioned pre-commit provider, or select exact gates through its args configuration.
For an LLM, use /llms.txt for the map, /llms-full.txt for the complete corpus, or replace /docs/<path> with /raw/<path>.md for a single source document.
Verify the local knowledge layer
The Ask experience resolves exact, aliased, and ambiguous documentation questions from a content-addressed, hash-verified static artifact before considering the optional backend. Run the same protocol verification locally:
node examples/verify-playbook.mjs
import { readFileSync } from 'node:fs'
import { decodeDeterministicSiteConfig, verifyLocalKnowledgeArtifactSync } from '@agentskit/chat/protocol'
const config = JSON.parse(readFileSync(new URL('../public/deterministic/site-config.json', import.meta.url)))
const artifact = JSON.parse(readFileSync(new URL('../public/deterministic/knowledge.json', import.meta.url)))
const site = decodeDeterministicSiteConfig(config)
if (!site.ok) throw new Error(site.diagnostic.message)
const verified = verifyLocalKnowledgeArtifactSync(artifact, {
expectedContentHash: site.value.artifact.contentHash,
expectedSiteId: site.value.siteId,
})
if (!verified.ok) throw new Error(verified.diagnostic.message)
console.log(`Verified ${verified.value.entries.length} local Playbook answers.`)
Expected output:
Verified 156 local Playbook answers.
How the system fits together
flowchart LR
S["Markdown source"] --> F["Fumadocs site"]
S --> D["Doc Bridge index"]
S --> L["llms.txt + raw routes"]
S --> K["Deterministic knowledge artifact"]
K --> A["AgentsKit Chat"]
A -->|"unresolved only"| B["Optional backend"]
S --> G["Executable quality gates"]
- Humans get a searchable Fumadocs journey, visual explanations, examples, and contribution paths.
- LLMs get stable raw Markdown, full-corpus exports, explicit ownership, and machine-readable local answers.
- Maintainers get generated statistics, Doc Bridge routing, README certification, tests, and build gates.
- Visitors can use Ask immediately for deterministic questions; the optional backend (set with
NEXT_PUBLIC_ASK_ENDPOINT) is called only when local evidence cannot answer safely.
See Discovery and Ask for the decision path, integrity checks, privacy behavior, and failure modes.
Repository map
Quality commands
pnpm test
pnpm test:playbook-package
pnpm check:playbook-package
pnpm check:okf-type
pnpm check:doc-bridge-config
pnpm docs:bridge:index
pnpm docs:bridge:gate
pnpm check:readme-standard
pnpm lint
pnpm build
The corpus is usable as versioned documentation today. The web application is currently 0.1.x; consumers should treat internal React components as private and depend on the published document and machine routes instead.
Contribute a production lesson
Contributions are welcome when a pattern is grounded in a real failure mode, explains its enforcement, and remains useful across tools. Read CONTRIBUTING.md, then use the contribution guide to validate structure, cross-links, machine retrieval, and gates.
Software contributions are licensed under MIT. Contributions to
the Playbook corpus—written patterns, prompts, templates, illustrations, and
other editorial content—are licensed under
CC BY 4.0. See governance, the
Code of Conduct, and the security policy
before contributing. Report vulnerabilities privately through the security
policy rather than a public issue.
AgentsKit ecosystem
Agents Playbook is the open engineering harness in the AgentsKit ecosystem:
- AgentsKit — build agents without gluing many libraries together.
- Registry — copy ready-made agents and own the source.
- AgentsKit Chat — define one conversational experience across interfaces.
- Doc Bridge — turn repository docs into executable agent handoffs.
- Code Review — configurable, evidence-based reviews before merge.
- Harness — configure the issue-to-release SDLC loop.
- Agents Playbook — this site's direct documentation entry point.
Topics: ai-agents · coding-agents · agent-governance · software-architecture · quality-gates · fumadocs · llms-txt · developer-experience
