OpenWriterControle/.opencode/context/CODEBASE_STANDARDS.md

OpenCode Codebase Standards

Complete Reference Guide
Last Updated: Feb 2026
Analyzed: 206+ TypeScript files across packages/opencode/src/

---

Table of Contents

1. Function Definition Standards
2. Class Usage Standards
3. Array Handling Standards
4. Variable & Destructuring Standards
5. Control Flow Standards
6. Async & Concurrency Standards
7. Race Condition Prevention
8. AI System Integration Standards
9. Service Architecture Standards
10. State Management Standards
11. Event Bus Standards
12. Configuration Management Standards
13. Storage & Persistence Standards
14. Error Handling Standards
15. Type System Standards
16. Import Organization Standards
17. Naming Conventions
18. Testing Standards
19. Documentation Standards
20. Schema Definition Standards
21. Dependency Management Standards
22. Build & Development Standards
23. Performance Standards
24. Security Standards

---

1. Function Definition Standards

1.1 Naming Convention

Rule: Prefer single-word function names

Compliance: 95%+

// ✅ GOOD - Single-word names
export const create = fn(...)
export const fork = fn(...)
export const touch = fn(...)
export const get = fn(...)
export const share = fn(...)
export async function stream(input: StreamInput) {...}
export async function work<T>(concurrency, items, fn) {...}

// ✅ ACCEPTABLE - Multi-word only when necessary
export function isDefaultTitle(title: string) {...}      // Boolean predicate
export function assertNotBusy(sessionID: string) {...}   // Assertion pattern
export async function createNext(input) {...}            // Version disambiguation
export async function resolvePromptParts(template) {...} // Complex operation needs clarity

// ❌ AVOID - Unnecessary multi-word names
function prepareJournal(dir: string) {}  // Use: journal()
function getUserData(id: string) {}      // Use: user()
function processFileContent(path) {}     // Use: process()

File References:

• packages/opencode/src/session/index.ts - Lines 140, 200, 206
• packages/opencode/src/session/llm.ts - Line 59
• packages/opencode/src/util/queue.ts - Lines 1-32

1.2 Function Definition Patterns

Pattern A: Zod-Validated Functions (Primary Pattern)

// packages/opencode/src/util/fn.ts - The fn() wrapper
export function fn<T extends z.ZodType, Result>(schema: T, cb: (input: z.infer<T>) => Result) {
  const result = (input: z.infer<T>) => {
    const parsed = schema.parse(input) // Validates input
    return cb(parsed)
  }
  result.force = (input: z.infer<T>) => cb(input) // Skip validation
  result.schema = schema // Expose schema
  return result
}

// Usage Example 1: Session creation
export const create = fn(
  z
    .object({
      parentID: Identifier.schema("session").optional(),
      title: z.string().optional(),
      permission: Info.shape.permission,
    })
    .optional(),
  async (input) => {
    return createNext({
      parentID: input?.parentID,
      title: input?.title,
      permission: input?.permission ?? "allow",
    })
  },
)

// Usage Example 2: Simple validation
export const touch = fn(Identifier.schema("session"), async (sessionID) => {
  await update(sessionID, (draft) => {
    draft.time.updated = Date.now()
  })
})

// Usage Example 3: Complex object validation
export const messages = fn(
  z.object({
    sessionID: Identifier.schema("session"),
    limit: z.number().optional(),
  }),
  async (input) => {
    const result = [] as MessageV2.WithParts[]
    for await (const msg of MessageV2.stream(input.sessionID)) {
      if (input.limit && result.length >= input.limit) break
      result.push(msg)
    }
    result.reverse()
    return result
  },
)

File References:

• packages/opencode/src/util/fn.ts - Lines 1-12
• packages/opencode/src/session/index.ts - Lines 140-202

Pattern B: Namespace Organization (Preferred over Classes)

// packages/opencode/src/session/index.ts
export namespace Session {
  const log = Log.create({ service: "session" })

  // Exported functions with fn() wrapper
  export const create = fn(...)
  export const fork = fn(...)
  export const touch = fn(...)
  export const messages = fn(...)

  // Async functions without wrapper
  export async function createNext(input) {...}
  export async function update(id, editor, options) {...}

  // Type definitions
  export type Info = z.infer<typeof Info>
  export const Info = z.object({...})

  // Events
  export const Event = {
    Created: BusEvent.define("session.created", z.object({info: Info})),
    Updated: BusEvent.define("session.updated", z.object({info: Info})),
  }

  // Error classes (only exception to no-class rule)
  export class BusyError extends Error {
    constructor(public readonly sessionID: string) {
      super(`Session ${sessionID} is busy`)
    }
  }
}

// Usage
await Session.create({ title: "New Session" })
await Session.touch(sessionID)
const msgs = await Session.messages({ sessionID, limit: 10 })

File References:

• packages/opencode/src/session/index.ts - Lines 26-518
• packages/opencode/src/tool/tool.ts - Lines 1-90
• packages/opencode/src/session/processor.ts - Lines 19-416

Pattern C: Inline vs. Extraction Rule

Rule: Inline when used once, extract when composable/reusable

// ✅ GOOD - Inline when value used once
const journal = await Bun.file(path.join(dir, "journal.json")).json()
const [language, cfg] = await Promise.all([Provider.getLanguage(input.model), Config.get()])

// ❌ BAD - Unnecessary intermediate variable
const journalPath = path.join(dir, "journal.json")
const journalFile = Bun.file(journalPath)
const journal = await journalFile.json()

// ✅ GOOD - Extract when reusable across multiple call sites
async function resolveTools(input: Pick<StreamInput, "tools" | "agent" | "user">) {
  const disabled = PermissionNext.disabled(Object.keys(input.tools), input.agent.permission)
  for (const tool of Object.keys(input.tools)) {
    if (input.user.tools?.[tool] === false || disabled.has(tool)) {
      delete input.tools[tool]
    }
  }
  return input.tools
}

// Called from multiple locations
const tools = await resolveTools({ tools: input.tools, agent, user })

// ✅ GOOD - Extract when logic is complex and improves readability
function shouldUseCopilotResponsesApi(modelID: string): boolean {
  const match = /^gpt-(\d+)/.exec(modelID)
  if (!match) return false
  return Number(match[1]) >= 5 && !modelID.startsWith("gpt-5-mini")
}

File References:

• packages/opencode/src/session/llm.ts - Lines 59, 268-276
• packages/opencode/src/provider/provider.ts - Lines 46-56

1.3 Function Composition Patterns

// ✅ GOOD - Functional composition with pipes
const filtered = agents
  .filter((a) => a.mode !== "primary")
  .filter((a) => PermissionNext.evaluate("task", a.name, caller.permission).action !== "deny")
  .map((a) => a.name)

// ✅ GOOD - Async composition with Promise.all
const [results, metadata] = await Promise.all([processItems(items), fetchMetadata(id)])

// ✅ GOOD - Higher-order functions
export function withLock<T>(filepath: string, fn: () => Promise<T>): Promise<T> {
  return Lock.write(filepath).then(async (lock) => {
    try {
      return await fn()
    } finally {
      lock[Symbol.dispose]()
    }
  })
}

---

2. Class Usage Standards

2.1 Avoid Classes (Except Specific Patterns)

Finding: Only 5 classes in 206 files

Rule: Use namespaces + functions instead of classes

// ❌ AVOID - Class-based organization
class SessionManager {
  private sessions: Map<string, Session> = new Map()

  create(input: CreateInput) {
    const session = { id: generateId(), ...input }
    this.sessions.set(session.id, session)
    return session
  }

  update(id: string, data: Partial<Session>) {
    const session = this.sessions.get(id)
    if (!session) throw new Error("Not found")
    Object.assign(session, data)
    return session
  }
}

// ✅ PREFER - Namespace-based organization
export namespace Session {
  const state = Instance.state(async () => ({
    sessions: new Map<string, Info>()
  }))

  export const create = fn(
    CreateInput.schema,
    async (input) => {
      const session = { id: Identifier.ascending("session"), ...input }
      state().sessions.set(session.id, session)
      return session
    }
  )

  export const update = fn(
    z.object({ id: z.string(), data: z.record(z.unknown()) }),
    async (input) => {
      const session = state().sessions.get(input.id)
      if (!session) throw NotFoundError.create({...})
      Object.assign(session, input.data)
      return session
    }
  )
}

2.2 Allowed Class Use Cases

Use Case 1: Error Types

// ✅ GOOD - Custom error classes
export class BusyError extends Error {
  constructor(public readonly sessionID: string) {
    super(`Session ${sessionID} is busy`)
  }
}

// Usage
throw new BusyError(sessionID)

// Catching
try {
  await operation()
} catch (error) {
  if (error instanceof BusyError) {
    // Handle busy state
  }
}

File Reference: packages/opencode/src/session/index.ts:494

Use Case 2: Protocol Implementations

// ✅ GOOD - Implementing external interfaces
export class OpenAICompatibleChatLanguageModel implements LanguageModelV2 {
  readonly specificationVersion = "v2"
  readonly provider: string
  readonly modelId: string

  constructor(modelId: string, settings: ModelSettings) {
    this.modelId = modelId
    this.provider = settings.provider
  }

  async doGenerate(options: GenerateOptions): Promise<GenerateResult> {
    // Implementation required by interface
  }

  async doStream(options: StreamOptions): AsyncIterableIterator<StreamPart> {
    // Implementation required by interface
  }
}

// ✅ GOOD - OAuth protocol implementation
export class McpOAuthProvider implements OAuthClientProvider {
  async getAccessToken(): Promise<string> {
    // OAuth flow implementation
  }
}

File References:

• packages/opencode/src/provider/sdk/* - Lines 53, 131
• packages/opencode/src/mcp/oauth-provider.ts:26

Use Case 3: Async Iterators

// ✅ GOOD - AsyncIterable implementation
export class AsyncQueue<T> implements AsyncIterable<T> {
  private queue: T[] = []
  private resolvers: ((value: T) => void)[] = []

  push(item: T) {
    const resolve = this.resolvers.shift()
    if (resolve) {
      resolve(item)
    } else {
      this.queue.push(item)
    }
  }

  async next(): Promise<T> {
    if (this.queue.length > 0) {
      return this.queue.shift()!
    }
    return new Promise((resolve) => {
      this.resolvers.push(resolve)
    })
  }

  async *[Symbol.asyncIterator]() {
    while (true) {
      yield await this.next()
    }
  }
}

// Usage
const queue = new AsyncQueue<Message>()
queue.push(message)

for await (const msg of queue) {
  console.log(msg)
}

File Reference: packages/opencode/src/util/queue.ts:1-19

Use Case 4: Complex Stateful Managers

// ✅ ACCEPTABLE - When managing complex state machines
export class ACPSessionManager {
  private sessions: Map<string, ACPSession>
  private connections: Map<string, WebSocket>

  // Complex lifecycle management
  async createSession(id: string) {...}
  async handleMessage(id: string, msg: Message) {...}
  async closeSession(id: string) {...}

  // State machine logic
  private transition(from: State, to: State) {...}
}

File Reference: packages/opencode/src/acp/session.ts:8

---

3. Array Handling Standards

3.1 Functional Methods (85% of operations)

Rule: Prefer map/filter/flatMap over for-loops

// ✅ GOOD - Functional chain with type inference
const files = messages
  .flatMap((x) => x.parts)
  .filter((x): x is Patch => x.type === "patch") // Type guard maintains inference
  .flatMap((x) => x.files)
  .map((x) => path.relative(Instance.worktree, x))

// ✅ GOOD - Parallel async operations
const results = await Promise.all(
  toolCalls.map(async (call) => {
    return executeCall(call)
  }),
)

// ✅ GOOD - Reduce for aggregation
const totalAdditions = diffs.reduce((sum, x) => sum + x.additions, 0)

// ✅ GOOD - Filter with type guards
const agents = await Agent.list().then((x) => x.filter((a): a is Agent & { mode: "secondary" } => a.mode !== "primary"))

// ✅ GOOD - Unique values
const uniqueNames = Array.from(new Set(items.map((x) => x.name)))

// ✅ GOOD - Sorting
const sorted = items.toSorted((a, b) => a.timestamp - b.timestamp)

File References:

• packages/opencode/src/tool/batch.ts - Lines 41, 56, 80, 170
• packages/opencode/src/session/prompt.ts - Lines 192-200
• packages/opencode/src/session/summary.ts - Lines 96-109

3.2 For-Loops (15%, only when necessary)

Rule: Use for-loops only for:

1. Algorithm complexity (DP, graph traversal)
2. Early exit requirements
3. Sequential side effects
4. Performance-critical iteration

Pattern 1: Algorithm Complexity

// ✅ GOOD - Dynamic programming (LCS algorithm)
// packages/opencode/src/tool/edit.ts:175-176
function lcs(a: string[], b: string[]): number[][] {
  const dp: number[][] = Array(a.length + 1)
    .fill(0)
    .map(() => Array(b.length + 1).fill(0))

  for (let i = 1; i <= a.length; i++) {
    for (let j = 1; j <= b.length; j++) {
      if (a[i - 1] === b[j - 1]) {
        dp[i][j] = dp[i - 1][j - 1] + 1
      } else {
        dp[i][j] = Math.max(dp[i - 1][j], dp[i][j - 1])
      }
    }
  }

  return dp
}

Pattern 2: Early Exit

// ✅ GOOD - Break on condition
// packages/opencode/src/session/revert.ts:31-40
const patches = []
for (const msg of all) {
  if (msg.info.id === revert.messageID) break
  for (const part of msg.parts) {
    if (part.type === "patch") {
      patches.push(part)
    }
  }
}

// ✅ GOOD - Find first match
for (const file of files) {
  if (await Filesystem.exists(file)) {
    return file
  }
}

Pattern 3: Sequential Side Effects

// ✅ GOOD - Sequential mutations
for (const key of Object.keys(input.tools)) {
  if (input.user.tools?.[key] === false || disabled.has(key)) {
    delete input.tools[key]
  }
}

// ✅ GOOD - Streaming output
for (const line of lines) {
  await stream.write(line)
}

File Reference: packages/opencode/src/tool/grep.ts:77

Pattern 4: Performance Critical

// ✅ GOOD - Low-level iteration for performance
for (let i = 0; i < buffer.length; i++) {
  result += buffer[i] * multiplier
}

// Note: Profile before optimizing - functional methods are often fast enough

3.3 Type Guards on Filter

Rule: Use type guards to maintain type inference downstream

// ✅ GOOD - Type guard preserves type information
const patches = messages.flatMap((msg) => msg.parts).filter((part): part is PatchPart => part.type === "patch")
// patches is now PatchPart[], not Part[]

// ❌ BAD - Loses type information
const patches = messages.flatMap((msg) => msg.parts).filter((part) => part.type === "patch")
// patches is still Part[], requires casting later

// ✅ GOOD - Multiple type guards
const validAgents = agents
  .filter((a): a is Agent => a !== undefined)
  .filter((a): a is Agent & { mode: "secondary" } => a.mode !== "primary")

File Reference: packages/opencode/src/tool/task.ts:24,29

---

4. Variable & Destructuring Standards

4.1 Variable Declaration

Rule: Prefer const over let

// ✅ GOOD - Immutable with ternary
const foo = condition ? 1 : 2
const result = await (isValid ? processValid() : processInvalid())

// ❌ BAD - Reassignment
let foo
if (condition) {
  foo = 1
} else {
  foo = 2
}

// ✅ GOOD - Early return instead of reassignment
function getValue(condition: boolean) {
  if (condition) return 1
  return 2
}

// ✅ ACCEPTABLE - let when mutation is necessary
let accumulator = 0
for (const item of items) {
  accumulator += item.value
}

File Reference: AGENTS.md:56-68

4.2 Destructuring

Rule: Avoid unnecessary destructuring, preserve context with dot notation

// ✅ GOOD - Preserve context
function process(session: Session) {
  log.info("processing", { id: session.id, title: session.title })
  return {
    id: session.id,
    status: session.status,
    owner: session.owner
  }
}

// ❌ BAD - Loses context, harder to read
function process(session: Session) {
  const { id, title, status, owner } = session
  log.info("processing", { id, title })
  return { id, status, owner }
}

// ✅ ACCEPTABLE - Destructuring when improving readability
function renderUser({ name, email, avatar }: User) {
  return `<div>${name} (${email})</div>`
}

// ✅ ACCEPTABLE - Destructuring array returns
const [language, cfg, provider] = await Promise.all([...])

File Reference: AGENTS.md:43-54

4.3 Variable Naming

Rule: Concise single-word names when descriptive

// ✅ GOOD
const session = await Session.get(id)
const user = await Auth.current()
const messages = await Session.messages({ sessionID })

// ❌ BAD - Unnecessary verbosity
const currentSession = await Session.get(id)
const currentlyAuthenticatedUser = await Auth.current()
const sessionMessagesList = await Session.messages({ sessionID })

// ✅ GOOD - Multi-word when single word is ambiguous
const sessionID = params.id
const userAgent = req.headers["user-agent"]
const maxRetries = config.retries

---

5. Control Flow Standards

5.1 Early Returns

Rule: Avoid else statements, use early returns

// ✅ GOOD - Early returns
function getStatus(session: Session) {
  if (!session) return "not_found"
  if (session.busy) return "busy"
  if (session.error) return "error"
  return "ready"
}

async function process(id: string) {
  const session = await Session.get(id)
  if (!session) return { error: "Not found" }

  const result = await execute(session)
  if (!result.success) return { error: result.message }

  return { data: result.data }
}

// ❌ BAD - Else statements
function getStatus(session: Session) {
  if (!session) {
    return "not_found"
  } else {
    if (session.busy) {
      return "busy"
    } else {
      if (session.error) {
        return "error"
      } else {
        return "ready"
      }
    }
  }
}

File Reference: AGENTS.md:70-86

5.2 Guard Clauses

// ✅ GOOD - Guard clauses at function start
async function updateSession(id: string, data: UpdateData) {
  if (!id) throw new Error("ID required")
  if (!data) throw new Error("Data required")
  if (data.title && data.title.length > 100) throw new Error("Title too long")

  // Main logic here
  const session = await Session.get(id)
  await Session.update(id, data)
  return session
}

5.3 Switch Statements

Rule: Use exhaustive switch with default case

// ✅ GOOD - Exhaustive switch for stream handling
for await (const value of stream.fullStream) {
  switch (value.type) {
    case "reasoning-start":
      reasoningMap[value.id] = { id: generateId(), type: "reasoning", text: "" }
      break

    case "reasoning-delta":
      if (value.id in reasoningMap) {
        reasoningMap[value.id].text += value.text
        await Session.updatePart({ delta: value.text })
      }
      break

    case "tool-call":
      await Session.updatePart({ state: { status: "running" } })
      break

    case "tool-result":
      await Session.updatePart({ state: { status: "completed" } })
      break

    default:
      const _exhaustive: never = value
      throw new Error(`Unhandled type: ${(value as any).type}`)
  }
}

File Reference: packages/opencode/src/session/processor.ts:53-346

---

6. Async & Concurrency Standards

6.1 Parallel Execution (Default Pattern)

Rule: Use Promise.all for independent operations

// ✅ GOOD - Parallel independent operations
const [language, cfg, provider, auth] = await Promise.all([
  Provider.getLanguage(input.model),
  Config.get(),
  Provider.getProvider(input.model.providerID),
  Auth.get(input.model.providerID),
])

// ✅ GOOD - Parallel array processing
const results = await Promise.all(
  items.map(async (item) => {
    return processItem(item)
  }),
)

// ❌ BAD - Sequential when independent
const language = await Provider.getLanguage(input.model)
const cfg = await Config.get() // Could run in parallel!
const provider = await Provider.getProvider(input.model.providerID)
const auth = await Auth.get(input.model.providerID)

File Reference: packages/opencode/src/session/llm.ts:59

6.2 Sequential Operations

Rule: Chain when operations depend on previous results

// ✅ GOOD - Sequential dependency chain
const session = await Session.create({ title: "New" })
const message = await Session.addMessage(session.id, { content: "Hello" })
const response = await LLM.stream({ sessionID: session.id, messageID: message.id })

// ✅ GOOD - Promise chain for clarity
const result = await Session.create({ title: "New" })
  .then((session) => Session.addMessage(session.id, { content: "Hello" }))
  .then((message) => LLM.stream({ messageID: message.id }))

6.3 Error Handling in Async

Rule: Prefer .catch() over try/catch when possible

// ✅ GOOD - Catch at call site
const result = await operation().catch((error) => {
  log.error("Operation failed", { error })
  return defaultValue
})

// ✅ GOOD - Promise.all with error handling
const results = await Promise.all(
  items.map(async (item) => {
    return processItem(item).catch((error) => {
      log.error("Item failed", { item, error })
      return null
    })
  }),
)

// ✅ ACCEPTABLE - try/catch for multiple operations
try {
  const session = await Session.create(input)
  await Session.addMessage(session.id, message)
  await EventBus.publish(Session.Event.Created, { session })
  return session
} catch (error) {
  log.error("Session creation failed", { error })
  throw error
}

// ❌ AVOID - try/catch for single operation
try {
  const result = await operation()
  return result
} catch (error) {
  log.error(error)
  throw error
}
// Better:
const result = await operation().catch((error) => {
  log.error(error)
  throw error
})

6.4 Concurrent Worker Pattern

// ✅ GOOD - Controlled concurrency
export async function work<T>(concurrency: number, items: T[], fn: (item: T) => Promise<void>) {
  const pending = [...items]

  await Promise.all(
    Array.from({ length: concurrency }, async () => {
      while (true) {
        const item = pending.pop()
        if (item === undefined) return
        await fn(item)
      }
    }),
  )
}

// Usage
await work(3, files, async (file) => {
  await processFile(file)
})

File Reference: packages/opencode/src/util/queue.ts:21-32

---

7. Race Condition Prevention

7.1 Reader-Writer Lock (Storage Operations)

// packages/opencode/src/util/lock.ts
export namespace Lock {
  const locks = new Map<
    string,
    {
      readers: number
      writer: boolean
      waitingReaders: (() => void)[]
      waitingWriters: (() => void)[]
    }
  >()

  export async function read(key: string): Promise<Disposable> {
    const lock = get(key)
    return new Promise((resolve) => {
      // Writers get priority
      if (!lock.writer && lock.waitingWriters.length === 0) {
        lock.readers++
        resolve({
          [Symbol.dispose]: () => {
            lock.readers--
            process(key)
          },
        })
      } else {
        lock.waitingReaders.push(() => {
          lock.readers++
          resolve({
            [Symbol.dispose]: () => {
              lock.readers--
              process(key)
            },
          })
        })
      }
    })
  }

  export async function write(key: string): Promise<Disposable> {
    const lock = get(key)
    return new Promise((resolve) => {
      if (!lock.writer && lock.readers === 0) {
        lock.writer = true
        resolve({
          [Symbol.dispose]: () => {
            lock.writer = false
            process(key)
          },
        })
      } else {
        lock.waitingWriters.push(() => {
          lock.writer = true
          resolve({
            [Symbol.dispose]: () => {
              lock.writer = false
              process(key)
            },
          })
        })
      }
    })
  }
}

// Usage with disposable pattern
export async function read<T>(target: string): Promise<T> {
  using _ = await Lock.read(target) // Auto-releases on scope exit
  const file = Bun.file(target)
  if (!(await file.exists())) {
    throw NotFoundError.create({ message: `File not found: ${target}` })
  }
  return file.json()
}

export async function write(target: string, data: any): Promise<void> {
  using _ = await Lock.write(target) // Auto-releases on scope exit
  const dir = path.dirname(target)
  await fs.mkdir(dir, { recursive: true })
  await Bun.write(target, JSON.stringify(data, null, 2))
}

File References:

• packages/opencode/src/util/lock.ts - Lines 1-99
• packages/opencode/src/storage/storage.ts - Lines 173, 183, 195

7.2 File-Level Lock (Edit Operations)

// packages/opencode/src/file/time.ts
export async function withLock<T>(filepath: string, fn: () => Promise<T>): Promise<T> {
  const current = state()
  const currentLock = current.locks.get(filepath) ?? Promise.resolve()

  let release: () => void = () => {}
  const nextLock = new Promise<void>((resolve) => {
    release = resolve
  })

  const chained = currentLock.then(() => nextLock)
  current.locks.set(filepath, chained)

  await currentLock // Wait for previous lock

  try {
    return await fn()
  } finally {
    release()
    if (current.locks.get(filepath) === chained) {
      current.locks.delete(filepath)
    }
  }
}

// Usage: Atomic read-modify-write
await FileTime.withLock(filePath, async () => {
  const contentOld = await Bun.file(filePath).text()
  const contentNew = replace(contentOld, params.oldString, params.newString)
  await Bun.write(filePath, contentNew)
  await FileTime.touch(filePath)
})

File References:

• packages/opencode/src/file/time.ts - Lines 35-53
• packages/opencode/src/tool/edit.ts - Line 50

7.3 Mutex Pattern

// ✅ GOOD - Simple mutex for critical sections
const mutexes = new Map<string, Promise<void>>()

async function withMutex<T>(key: string, fn: () => Promise<T>): Promise<T> {
  const current = mutexes.get(key) ?? Promise.resolve()

  let release: () => void
  const next = new Promise<void>((resolve) => {
    release = resolve
  })
  mutexes.set(key, next)

  await current

  try {
    return await fn()
  } finally {
    release!()
    if (mutexes.get(key) === next) {
      mutexes.delete(key)
    }
  }
}

// Usage
await withMutex(`session:${sessionID}`, async () => {
  // Critical section - only one execution at a time
  const session = await Session.get(sessionID)
  session.busy = true
  await Session.save(session)
})

7.4 Debouncing & Throttling

// ✅ GOOD - Debounce for file watching
function debounce<T extends (...args: any[]) => any>(fn: T, delay: number): (...args: Parameters<T>) => void {
  let timer: Timer | undefined

  return (...args: Parameters<T>) => {
    if (timer) clearTimeout(timer)
    timer = setTimeout(() => {
      fn(...args)
      timer = undefined
    }, delay)
  }
}

// Usage in file watcher
const debouncedUpdate = debounce((file: string) => {
  Bus.publish(FileWatcher.Event.Updated, { file })
}, 100)

watcher.on("change", (file) => {
  debouncedUpdate(file)
})

---

8. AI System Integration Standards

8.1 Message Context Management

// packages/opencode/src/session/message-v2.ts

export type Part = TextPart | FilePart | PatchPart | ToolCallPart | ToolResultPart | ReasoningPart

export interface WithParts {
  info: Info
  parts: Part[]
}

// Convert internal message format to LLM format
export function toModelMessages(input: WithParts[], model: Provider.Model): ModelMessage[] {
  return input.flatMap((msg): ModelMessage[] => {
    if (msg.info.role === "user") {
      return [
        {
          role: "user",
          content: msg.parts.map((part) => {
            if (part.type === "text") {
              return { type: "text", text: part.text }
            }
            if (part.type === "file") {
              return {
                type: "file",
                data: part.data,
                mimeType: part.mimeType,
              }
            }
            if (part.type === "patch") {
              return {
                type: "text",
                text: formatPatch(part),
              }
            }
            // ... handle other types
          }),
        },
      ]
    }

    if (msg.info.role === "assistant") {
      return [
        {
          role: "assistant",
          content: msg.parts
            .filter((p) => p.type === "text" || p.type === "reasoning")
            .map((p) => ({ type: "text", text: p.text })),
        },
      ]
    }

    // Handle tool results...
    return []
  })
}

File Reference: packages/opencode/src/session/message-v2.ts:438-500

8.2 System Prompt Construction

// packages/opencode/src/session/llm.ts

export async function stream(input: StreamInput) {
  const system: string[] = []

  // Part 1: Base system prompt (cached)
  const header = [
    // Agent prompt OR provider prompt
    ...(input.agent.prompt ? [input.agent.prompt] : isCodex ? [] : SystemPrompt.provider(input.model)),
    // Custom prompt from request
    ...input.system,
    // Custom prompt from user config
    ...(input.user.system ? [input.user.system] : []),
  ]
    .filter(Boolean)
    .join("\n")

  system.push(header)

  // Part 2: Dynamic context (not cached)
  const dynamicContext = [
    `Working directory: ${Instance.directory}`,
    `Current date: ${new Date().toISOString()}`,
    // ... other dynamic info
  ].join("\n")

  system.push(dynamicContext)

  // Allow plugins to transform system prompt
  const original = clone(system)
  await Plugin.trigger(
    "experimental.chat.system.transform",
    {
      model: input.model,
      agent: input.agent,
    },
    { system },
  )

  // Optimize for prompt caching: maintain 2-part structure if header unchanged
  if (system.length > 2 && system[0] === header) {
    const rest = system.slice(1)
    system.length = 0
    system.push(header, rest.join("\n"))
  }

  // Convert to LLM messages
  const messages = [
    ...system.map((text) => ({ role: "system", content: text })),
    ...toModelMessages(input.messages, input.model),
  ]

  return language.doStream({
    model: input.model.id,
    messages,
    tools: await resolveTools(input),
  })
}

File Reference: packages/opencode/src/session/llm.ts:67-97

8.3 Tool Definition & Execution

// packages/opencode/src/tool/tool.ts

export namespace Tool {
  export interface Info<Parameters = any, Metadata = any> {
    id: string
    title: string
    description: string
    parameters: z.ZodType<Parameters>
    metadata?: Metadata

    execute(args: Parameters, context: Context): Promise<ExecuteResult>
  }

  export interface Context {
    sessionID: string
    messageID: string
    agent: string
    abort: AbortSignal
    messages: () => Promise<MessageV2.WithParts[]>
    metadata: <T>(key: string, value: T) => Promise<void>
    ask: <T>(question: Question<T>) => Promise<T>
  }

  export type ExecuteResult = {
    output?: string
    title?: string
    metadata?: Record<string, unknown>
  }

  export function define<Parameters, Result>(
    id: string,
    init: {
      title: string
      description: string
      parameters: z.ZodType<Parameters>
      execute: (args: Parameters, context: Context) => Promise<ExecuteResult>
    },
  ): Info<Parameters> {
    return {
      id,
      ...init,
    }
  }
}

// Usage: Define a tool
export const ReadTool = Tool.define("read", {
  title: "Read File",
  description: "Read contents of a file from the filesystem",
  parameters: z.object({
    filePath: z.string().describe("Absolute path to file"),
    offset: z.number().optional().describe("Line number to start reading"),
    limit: z.number().optional().describe("Number of lines to read"),
  }),

  async execute(args, context) {
    await context.ask({
      type: "permission",
      message: `Read file ${args.filePath}?`,
      actions: ["allow", "deny"],
    })

    const content = await Bun.file(args.filePath).text()
    const lines = content.split("\n")

    const start = args.offset ?? 0
    const end = args.limit ? start + args.limit : lines.length
    const selected = lines.slice(start, end)

    return {
      output: selected.join("\n"),
      title: `Read ${path.basename(args.filePath)}`,
      metadata: {
        lineCount: selected.length,
        filePath: args.filePath,
      },
    }
  },
})

// Register tool
await ToolRegistry.register(ReadTool)

// Tool execution in session
const tools = await ToolRegistry.tools({ agent: input.agent })

const toolMap = tools.reduce(
  (acc, tool) => {
    acc[tool.id] = vercelAiTool({
      description: tool.description,
      parameters: zodToJsonSchema(tool.parameters),
      execute: async (args) => {
        const result = await tool.execute(args, {
          sessionID: input.sessionID,
          messageID: input.messageID,
          agent: input.agent.name,
          abort: input.abort,
          messages: () => Session.messages({ sessionID: input.sessionID }),
          metadata: async (key, value) => {
            await Session.updatePart({
              messageID: input.messageID,
              metadata: { [key]: value },
            })
          },
          ask: async (question) => {
            return new Promise((resolve) => {
              Bus.publish(Session.Event.Question, {
                sessionID: input.sessionID,
                question,
                respond: resolve,
              })
            })
          },
        })

        return result.output
      },
    })
    return acc
  },
  {} as Record<string, any>,
)

File References:

• packages/opencode/src/tool/tool.ts - Lines 1-90
• packages/opencode/src/session/prompt.ts - Lines 711-748

8.4 Streaming Handling

// packages/opencode/src/session/processor.ts

export function create(input: CreateInput) {
  const reasoningMap: Record<string, ReasoningPart> = {}
  const toolcalls: Record<string, ToolCallPart> = {}

  return {
    async process() {
      const stream = await LLM.stream({
        sessionID: input.sessionID,
        messageID: input.messageID,
        model: input.model,
        messages: input.messages,
        tools: input.tools,
        abort: input.abort,
      })

      // Handle stream events
      for await (const value of stream.fullStream) {
        input.abort.throwIfAborted()

        switch (value.type) {
          case "reasoning-start":
            reasoningMap[value.id] = {
              id: Identifier.ascending("part"),
              type: "reasoning",
              text: "",
              time: { start: Date.now() },
            }
            await Session.addPart({
              messageID: input.messageID,
              part: reasoningMap[value.id],
            })
            break

          case "reasoning-delta":
            if (value.id in reasoningMap) {
              const part = reasoningMap[value.id]
              part.text += value.text
              await Session.updatePart({
                messageID: input.messageID,
                partID: part.id,
                delta: value.text,
              })
            }
            break

          case "reasoning-finish":
            if (value.id in reasoningMap) {
              const part = reasoningMap[value.id]
              part.time.end = Date.now()
              await Session.updatePart({
                messageID: input.messageID,
                partID: part.id,
                part,
              })
            }
            break

          case "text-delta":
            // Handle text streaming...
            break

          case "tool-call":
            const toolPart: ToolCallPart = {
              id: Identifier.ascending("part"),
              type: "tool-call",
              toolCallID: value.toolCallId,
              toolName: value.toolName,
              state: {
                status: "running",
                input: value.args,
                time: { start: Date.now() },
              },
            }
            toolcalls[value.toolCallId] = toolPart
            await Session.addPart({
              messageID: input.messageID,
              part: toolPart,
            })
            break

          case "tool-result":
            const match = toolcalls[value.toolCallId]
            if (match) {
              match.state = {
                status: "completed",
                output: value.result,
                time: {
                  start: match.state.time.start,
                  end: Date.now(),
                },
              }
              await Session.updatePart({
                messageID: input.messageID,
                partID: match.id,
                part: match,
              })
            }
            break

          case "error":
            await Session.updateMessage({
              messageID: input.messageID,
              error: {
                message: value.error.message,
                code: value.error.code,
              },
            })
            throw value.error

          case "finish":
            await Session.updateMessage({
              messageID: input.messageID,
              tokens: value.usage,
              time: { end: Date.now() },
            })
            break

          default:
            // Exhaustiveness check
            const _exhaustive: never = value
            throw new Error(`Unhandled stream type: ${(value as any).type}`)
        }
      }
    },
  }
}

File Reference: packages/opencode/src/session/processor.ts:53-346

8.5 Context Compaction

// packages/opencode/src/session/compaction.ts

export async function isOverflow(input: {
  tokens: MessageV2.Assistant["tokens"]
  model: Provider.Model
}): Promise<boolean> {
  const limit = input.model.limit.context

  const total = input.tokens.input + input.tokens.output + input.tokens.cache.read + input.tokens.cache.write

  const threshold = limit * 0.75 // Trigger at 75% capacity

  return total > threshold
}

export async function compact(sessionID: string) {
  const messages = await Session.messages({ sessionID })
  const model = await Session.getModel(sessionID)

  // Find oldest non-system messages
  const candidates = messages.filter((m) => m.info.role !== "system")

  if (candidates.length <= 2) {
    // Keep at least 2 messages for context
    return
  }

  // Remove oldest message
  const toRemove = candidates[0]
  await Session.deleteMessage(toRemove.info.id)

  // Add summary if necessary
  const summary = await generateSummary(toRemove)
  await Session.addSystemMessage(sessionID, {
    content: `Previous context (summarized): ${summary}`,
  })

  log.info("compacted session", {
    sessionID,
    removedMessageID: toRemove.info.id,
    remainingMessages: messages.length - 1,
  })
}

// Check after each assistant response
await Session.onMessageComplete(async (messageID) => {
  const message = await Session.getMessage(messageID)
  const model = await Session.getModel(message.sessionID)

  if (await isOverflow({ tokens: message.tokens, model })) {
    await compact(message.sessionID)
  }
})

File References:

• packages/opencode/src/session/compaction.ts - Lines 30-48
• packages/opencode/src/session/processor.ts - Lines 282-284

---

9. Service Architecture Standards

9.1 Core Services Overview

Service	Location	Pattern	Responsibility
Session Management	src/session/	Namespace + State	AI conversation lifecycle
File Watching	src/file/watcher.ts	Chokidar + Debounce	Filesystem change detection
LSP Integration	src/lsp/	Multi-client manager	Language server protocol
MCP	src/mcp/	Multi-transport client	Model Context Protocol
Terminal/PTY	src/server/routes/pty.ts	WebSocket-based	Shell session management
Event Bus	src/bus/	Pub/sub pattern	Cross-service communication
Storage	src/storage/	Lock-based persistence	JSON file storage
Provider Management	src/provider/	Lazy-loaded SDKs	LLM provider abstraction
Config Management	src/config/	Layered merge	Multi-level configuration
Scheduler	src/scheduler/	Interval-based	Background job execution

9.2 Service Implementation Template

// Service template following codebase patterns

import { Log } from "../util/log"
import { Instance } from "../project/instance"
import { BusEvent } from "../bus/event"
import z from "zod"

export namespace MyService {
  const log = Log.create({ service: "my-service" })

  // 1. Type definitions
  export type Info = z.infer<typeof Info>
  export const Info = z.object({
    id: z.string(),
    status: z.enum(["idle", "active", "error"]),
    created: z.number(),
  })

  // 2. Per-instance state
  const state = Instance.state(
    async () => {
      log.info("initializing service")

      return {
        items: new Map<string, Info>(),
        timers: new Map<string, Timer>(),
        clients: [] as Client[],
      }
    },
    async (state) => {
      log.info("disposing service")

      // Cleanup resources
      await Promise.all([
        ...state.clients.map((c) => c.close()),
        ...Array.from(state.timers.values()).map((t) => clearInterval(t)),
      ])

      state.items.clear()
    },
  )

  // 3. Event definitions
  export const Event = {
    Created: BusEvent.define(
      "my-service.created",
      z.object({
        info: Info,
      }),
    ),
    Updated: BusEvent.define(
      "my-service.updated",
      z.object({
        info: Info,
      }),
    ),
    Deleted: BusEvent.define(
      "my-service.deleted",
      z.object({
        id: z.string(),
      }),
    ),
  }

  // 4. Public API with fn() wrapper
  export const create = fn(
    z
      .object({
        id: z.string().optional(),
        data: z.record(z.unknown()).optional(),
      })
      .optional(),
    async (input) => {
      const id = input?.id ?? Identifier.ascending("my-service")

      const info: Info = {
        id,
        status: "idle",
        created: Date.now(),
      }

      state().items.set(id, info)
      await Bus.publish(Event.Created, { info })

      log.info("created", { id })
      return info
    },
  )

  export const get = fn(z.string(), async (id) => {
    const item = state().items.get(id)
    if (!item) {
      throw NotFoundError.create({ message: `Item ${id} not found` })
    }
    return item
  })

  export const list = fn(
    z
      .object({
        status: z.enum(["idle", "active", "error"]).optional(),
      })
      .optional(),
    async (input) => {
      const items = Array.from(state().items.values())

      if (input?.status) {
        return items.filter((item) => item.status === input.status)
      }

      return items
    },
  )

  // 5. Internal functions (not exported)
  async function cleanup(id: string) {
    const timer = state().timers.get(id)
    if (timer) {
      clearInterval(timer)
      state().timers.delete(id)
    }
  }
}

File References:

• packages/opencode/src/session/index.ts - Complete example
• packages/opencode/src/mcp/index.ts - Multi-client manager
• packages/opencode/src/lsp/index.ts - Lazy spawning

---

10. State Management Standards

10.1 Per-Instance State Isolation

Rule: All state must be isolated per working directory

// packages/opencode/src/project/instance.ts

export namespace Instance {
  // Current working directory
  export let directory = process.cwd()

  // Git worktree root
  export let worktree = directory

  // Project ID
  export let id = "unknown"

  // Create state scoped to current instance
  export function state<S>(
    init: () => S | Promise<S>,
    dispose?: (state: Awaited<S>) => Promise<void>,
  ): () => Awaited<S> {
    return State.create(
      () => Instance.directory, // Key by directory
      init,
      dispose,
    )
  }
}

// packages/opencode/src/project/state.ts

const recordsByKey = new Map<string, Map<Function, Entry>>()

interface Entry {
  state: any
  dispose?: (state: any) => Promise<void>
}

export function create<S>(
  root: () => string,
  init: () => S | Promise<S>,
  dispose?: (state: Awaited<S>) => Promise<void>,
): () => Awaited<S> {
  return () => {
    const key = root()

    // Get or create entries for this key
    let entries = recordsByKey.get(key)
    if (!entries) {
      entries = new Map<Function, Entry>()
      recordsByKey.set(key, entries)
    }

    // Get or initialize state
    const exists = entries.get(init)
    if (exists) {
      return exists.state as Awaited<S>
    }

    const state = init()
    entries.set(init, { state, dispose })

    return state as Awaited<S>
  }
}

// Dispose all state for a key
export async function dispose(key: string) {
  const entries = recordsByKey.get(key)
  if (!entries) return

  await Promise.all(
    Array.from(entries.values()).map(async (entry) => {
      if (entry.dispose) {
        const state = await entry.state
        await entry.dispose(state)
      }
    }),
  )

  recordsByKey.delete(key)
}

File References:

• packages/opencode/src/project/instance.ts - Lines 66-68
• packages/opencode/src/project/state.ts - Lines 12-29

10.2 State Initialization Pattern

// ✅ GOOD - Lazy initialization with disposal
const state = Instance.state(
  async () => {
    log.info("initializing LSP clients")

    const cfg = await Config.get()
    const servers: Record<string, LSPServer.Info> = {}
    const clients: LSPClient.Info[] = []

    // Initialize servers
    for (const [id, server] of Object.entries(LSPServer)) {
      if (cfg.lsp?.[id]?.enabled !== false) {
        servers[id] = server
      }
    }

    return {
      servers,
      clients,
      broken: new Set<string>(),
      spawning: new Map<string, Promise<LSPClient.Info | undefined>>(),
    }
  },
  async (state) => {
    log.info("disposing LSP clients")

    // Cleanup all clients
    await Promise.all(
      state.clients.map(async (client) => {
        try {
          await client.shutdown()
        } catch (error) {
          log.error("failed to shutdown client", { client: client.id, error })
        }
      }),
    )
  },
)

// Access state
const lspState = state()
lspState.clients.push(newClient)

10.3 Global State (Rare)

Rule: Avoid global state, use per-instance state instead

// ❌ AVOID - Global mutable state
const sessions = new Map<string, Session>()

export function addSession(session: Session) {
  sessions.set(session.id, session)
}

// ✅ PREFER - Per-instance state
const state = Instance.state(async () => ({
  sessions: new Map<string, Session>(),
}))

export function addSession(session: Session) {
  state().sessions.set(session.id, session)
}

// ✅ ACCEPTABLE - Truly global state (user config, auth tokens)
export namespace Global {
  export namespace Path {
    export const home = os.homedir()
    export const config = path.join(home, ".config", "opencode")
    export const data = path.join(home, ".local", "share", "opencode")
    export const cache = path.join(home, ".cache", "opencode")
  }
}

---

11. Event Bus Standards

11.1 Event Definition

// packages/opencode/src/bus/event.ts

export namespace BusEvent {
  export interface Definition<Properties extends z.ZodType = any> {
    type: string
    properties: Properties
  }

  export function define<Properties extends z.ZodType>(type: string, properties: Properties): Definition<Properties> {
    return { type, properties }
  }
}

// Usage: Define events for a service
export namespace Session {
  export const Event = {
    Created: BusEvent.define("session.created", z.object({ info: Info })),

    Updated: BusEvent.define("session.updated", z.object({ info: Info, changes: z.record(z.unknown()) })),

    Deleted: BusEvent.define("session.deleted", z.object({ id: z.string() })),

    Diff: BusEvent.define(
      "session.diff",
      z.object({
        sessionID: z.string(),
        messageID: z.string(),
        diff: z.object({
          file: z.string(),
          additions: z.number(),
          deletions: z.number(),
        }),
      }),
    ),

    Error: BusEvent.define(
      "session.error",
      z.object({
        sessionID: z.string(),
        error: z.object({
          message: z.string(),
          code: z.string().optional(),
        }),
      }),
    ),

    Question: BusEvent.define(
      "session.question",
      z.object({
        sessionID: z.string(),
        question: z.object({
          type: z.string(),
          message: z.string(),
          options: z.array(z.unknown()),
        }),
        respond: z.function(),
      }),
    ),
  }
}

11.2 Publishing Events

// packages/opencode/src/bus/index.ts

export namespace Bus {
  const state = Instance.state(async () => ({
    subscriptions: new Map<string, Set<Handler>>(),
  }))

  type Handler = (payload: EventPayload) => void | Promise<void>

  interface EventPayload {
    type: string
    properties: any
  }

  export async function publish<Definition extends BusEvent.Definition>(
    def: Definition,
    properties: z.output<Definition["properties"]>,
  ): Promise<void> {
    const payload: EventPayload = {
      type: def.type,
      properties,
    }

    const pending: Promise<void>[] = []

    // Notify specific subscribers
    const specific = state().subscriptions.get(def.type)
    if (specific) {
      for (const handler of specific) {
        pending.push(Promise.resolve(handler(payload)))
      }
    }

    // Notify wildcard subscribers
    const wildcard = state().subscriptions.get("*")
    if (wildcard) {
      for (const handler of wildcard) {
        pending.push(Promise.resolve(handler(payload)))
      }
    }

    // Also publish to global bus (for cross-instance communication)
    GlobalBus.emit("event", {
      directory: Instance.directory,
      payload,
    })

    await Promise.all(pending)
  }
}

// Usage: Publish an event
await Bus.publish(Session.Event.Created, {
  info: {
    id: "session-123",
    title: "New Session",
    created: Date.now(),
  },
})

await Bus.publish(Session.Event.Updated, {
  info: updatedSession,
  changes: { title: "Updated Title" },
})

File Reference: packages/opencode/src/bus/index.ts:41-64

11.3 Subscribing to Events

// packages/opencode/src/bus/index.ts

export namespace Bus {
  export function subscribe<Definition extends BusEvent.Definition>(
    def: Definition | "*",
    handler: (payload: { type: string; properties: z.output<Definition["properties"]> }) => void | Promise<void>,
  ): Disposable {
    const key = typeof def === "string" ? def : def.type

    let subs = state().subscriptions.get(key)
    if (!subs) {
      subs = new Set()
      state().subscriptions.set(key, subs)
    }

    subs.add(handler)

    return {
      [Symbol.dispose]: () => {
        subs?.delete(handler)
        if (subs?.size === 0) {
          state().subscriptions.delete(key)
        }
      },
    }
  }
}

// Usage: Subscribe to specific events
const subscription = Bus.subscribe(Session.Event.Created, async (event) => {
  log.info("session created", { id: event.properties.info.id })
  await notifyUser(`New session: ${event.properties.info.title}`)
})

// Unsubscribe
subscription[Symbol.dispose]()

// Or use disposable pattern
{
  using sub = Bus.subscribe(Session.Event.Updated, handleUpdate)

  // Automatically unsubscribes when scope exits
}

// Subscribe to all events
Bus.subscribe("*", (event) => {
  log.debug("event", { type: event.type, properties: event.properties })
})

11.4 Event-Driven Communication Pattern

// Service A: Publish events
export namespace FileWatcher {
  export const Event = {
    Updated: BusEvent.define(
      "file.updated",
      z.object({
        file: z.string(),
        event: z.enum(["add", "change", "unlink"]),
      }),
    ),
  }

  async function watch(directory: string) {
    const watcher = chokidar.watch(directory)

    watcher.on("change", async (file) => {
      await Bus.publish(Event.Updated, {
        file,
        event: "change",
      })
    })
  }
}

// Service B: React to events
export namespace LSP {
  export async function initialize() {
    // Subscribe to file changes
    Bus.subscribe(FileWatcher.Event.Updated, async (event) => {
      if (event.properties.event === "change") {
        await notifyClientsOfChange(event.properties.file)
      }
    })
  }

  async function notifyClientsOfChange(file: string) {
    const clients = state().clients.filter((c) => c.watchesFile(file))
    await Promise.all(clients.map((c) => c.didChangeTextDocument({ uri: file })))
  }
}

// Service C: Also react to same events
export namespace Session {
  export async function initialize() {
    Bus.subscribe(FileWatcher.Event.Updated, async (event) => {
      // Invalidate cached file contents
      await invalidateFileCache(event.properties.file)
    })
  }
}

---

12. Configuration Management Standards

12.1 Configuration Precedence

Order (Low → High priority):

1. Remote .well-known/opencode - Organization defaults
2. Global config - ~/.config/opencode/opencode.json{,c}
3. Custom config - OPENCODE_CONFIG env var path
4. Project config - opencode.json{,c} in project
5. .opencode directories - .opencode/opencode.json{,c}
6. Inline config - OPENCODE_CONFIG_CONTENT env var
7. Managed config - Enterprise /etc/opencode (highest priority)

// packages/opencode/src/config/config.ts

export namespace Config {
  export const state = Instance.state(async () => {
    let result: Info = {}

    // 1. Remote organization config
    const auth = await Auth.all()
    for (const [key, value] of Object.entries(auth)) {
      if (value.type === "wellknown") {
        const response = await fetch(`${key}/.well-known/opencode`)
        const wellknown = await response.json()
        result = mergeConfigConcatArrays(result, wellknown.config ?? {})
      }
    }

    // 2. Global user config
    result = mergeConfigConcatArrays(result, await global())

    // 3. Custom config path
    if (Flag.OPENCODE_CONFIG) {
      result = mergeConfigConcatArrays(result, await loadFile(Flag.OPENCODE_CONFIG))
    }

    // 4. Project config (if not disabled)
    if (!Flag.OPENCODE_DISABLE_PROJECT_CONFIG) {
      for (const file of ["opencode.jsonc", "opencode.json"]) {
        const found = await Filesystem.findUp(file, Instance.directory, Instance.worktree)
        for (const resolved of found.toReversed()) {
          result = mergeConfigConcatArrays(result, await loadFile(resolved))
        }
      }
    }

    // 5. .opencode directories
    const directories = [
      Global.Path.config,
      ...(!Flag.OPENCODE_DISABLE_PROJECT_CONFIG
        ? await Array.fromAsync(
            Filesystem.up({
              targets: [".opencode"],
              start: Instance.directory,
              stop: Instance.worktree,
            }),
          )
        : []),
      // Always scan user home ~/.opencode/
      ...(await Array.fromAsync(
        Filesystem.up({
          targets: [".opencode"],
          start: Global.Path.home,
          stop: Global.Path.home,
        }),
      )),
    ]

    for (const dir of unique(directories)) {
      for (const file of ["opencode.jsonc", "opencode.json"]) {
        const configPath = path.join(dir, file)
        if (await Filesystem.exists(configPath)) {
          result = mergeConfigConcatArrays(result, await loadFile(configPath))
        }
      }
    }

    // 6. Inline config
    if (Flag.OPENCODE_CONFIG_CONTENT) {
      result = mergeConfigConcatArrays(result, await load(Flag.OPENCODE_CONFIG_CONTENT, "inline"))
    }

    // 7. Managed config (enterprise, highest priority)
    const managedPath = path.join(managedConfigDir, "opencode.json")
    if (await Filesystem.exists(managedPath)) {
      result = mergeConfigConcatArrays(result, await loadFile(managedPath))
    }

    return result
  })

  export const get = fn(z.void(), async () => state())
}

File Reference: packages/opencode/src/config/config.ts:62-150

12.2 Configuration Merge Strategy

// packages/opencode/src/config/config.ts

function mergeConfigConcatArrays(target: Info, source: Info): Info {
  // Deep merge objects
  const merged = mergeDeep(target, source)

  // Special handling: Concatenate arrays instead of replacing
  if (target.plugin && source.plugin) {
    merged.plugin = Array.from(new Set([...target.plugin, ...source.plugin]))
  }

  if (target.instructions && source.instructions) {
    merged.instructions = Array.from(new Set([...target.instructions, ...source.instructions]))
  }

  return merged
}

// Example:
// Base config:    { plugin: ["a", "b"], model: { id: "gpt-4" } }
// Override config: { plugin: ["b", "c"], model: { temperature: 0.7 } }
// Result:         { plugin: ["a", "b", "c"], model: { id: "gpt-4", temperature: 0.7 } }

File Reference: packages/opencode/src/config/config.ts:51-60

12.3 Configuration Schema

export namespace Config {
  export type Info = z.infer<typeof Info>
  export const Info = z.object({
    // Model configuration
    model: z
      .object({
        id: z.string().optional(),
        providerID: z.string().optional(),
        temperature: z.number().optional(),
        maxTokens: z.number().optional(),
      })
      .optional(),

    // Providers
    providers: z
      .record(
        z.object({
          apiKey: z.string().optional(),
          baseURL: z.string().optional(),
          enabled: z.boolean().optional(),
        }),
      )
      .optional(),

    // Plugins
    plugin: z.array(z.string()).optional(),

    // Instructions (system prompts)
    instructions: z.array(z.string()).optional(),

    // Agents
    agent: z
      .record(
        z.object({
          mode: z.enum(["primary", "secondary"]).optional(),
          prompt: z.string().optional(),
          permission: z.string().optional(),
        }),
      )
      .optional(),

    // LSP servers
    lsp: z
      .record(
        z.object({
          enabled: z.boolean().optional(),
          command: z.string().optional(),
          args: z.array(z.string()).optional(),
        }),
      )
      .optional(),

    // MCP servers
    mcp: z
      .record(
        z.union([
          z.object({
            command: z.string(),
            args: z.array(z.string()).optional(),
            env: z.record(z.string()).optional(),
            enabled: z.boolean().optional(),
          }),
          z.object({
            url: z.string(),
            transport: z.literal("sse"),
            enabled: z.boolean().optional(),
          }),
        ]),
      )
      .optional(),

    // Formatters
    formatter: z
      .record(
        z.object({
          command: z.string(),
          args: z.array(z.string()).optional(),
        }),
      )
      .optional(),
  })
}

12.4 Configuration File Loading

export namespace Config {
  async function loadFile(filepath: string): Promise<Info> {
    if (!(await Filesystem.exists(filepath))) {
      return {}
    }

    const content = await Bun.file(filepath).text()
    return load(content, filepath)
  }

  async function load(content: string, source: string): Promise<Info> {
    try {
      // Parse JSONC (JSON with comments)
      const parsed = parseJsonc(content)

      // Validate against schema
      const validated = Info.parse(parsed)

      return validated
    } catch (error) {
      if (error instanceof z.ZodError) {
        log.error("config validation failed", {
          source,
          errors: error.errors,
        })
        throw new Error(`Invalid config at ${source}: ${error.errors[0].message}`)
      }

      throw error
    }
  }
}

---

13. Storage & Persistence Standards

13.1 Storage Operations

// packages/opencode/src/storage/storage.ts

export namespace Storage {
  const log = Log.create({ service: "storage" })

  export const NotFoundError = NamedError.create("NotFoundError", z.object({ message: z.string() }))

  // Read with lock
  export async function read<T>(target: string): Promise<T> {
    using _ = await Lock.read(target)

    const file = Bun.file(target)
    if (!(await file.exists())) {
      throw NotFoundError.create({
        message: `File not found: ${target}`,
      })
    }

    return file.json()
  }

  // Write with lock
  export async function write(target: string, data: any): Promise<void> {
    using _ = await Lock.write(target)

    const dir = path.dirname(target)
    await fs.mkdir(dir, { recursive: true })

    await Bun.write(target, JSON.stringify(data, null, 2))
  }

  // Update (read-modify-write)
  export async function update<T>(target: string, editor: (draft: T) => void | Promise<void>): Promise<T> {
    using _ = await Lock.write(target)

    const data = await read<T>(target)
    await editor(data)
    await write(target, data)

    return data
  }

  // Delete
  export async function remove(target: string): Promise<void> {
    using _ = await Lock.write(target)

    if (await Filesystem.exists(target)) {
      await fs.rm(target, { recursive: true })
    }
  }

  // List files matching pattern
  export async function list(directory: string, pattern: string): Promise<string[]> {
    using _ = await Lock.read(directory)

    const files: string[] = []
    for await (const file of new Bun.Glob(pattern).scan({ cwd: directory })) {
      files.push(path.join(directory, file))
    }

    return files
  }
}

File Reference: packages/opencode/src/storage/storage.ts:1-200

13.2 Storage Paths

export namespace Storage {
  export function path(...segments: string[]): string {
    return path.join(Global.Path.data, ...segments)
  }

  // Session storage
  export namespace Session {
    export function info(sessionID: string): string {
      return Storage.path("session", sessionID, "info.json")
    }

    export function messages(sessionID: string): string {
      return Storage.path("session", sessionID, "messages")
    }

    export function message(sessionID: string, messageID: string): string {
      return Storage.path("session", sessionID, "messages", `${messageID}.json`)
    }
  }

  // Project storage
  export namespace Project {
    export function info(projectID: string): string {
      return Storage.path("project", `${projectID}.json`)
    }

    export function list(): Promise<string[]> {
      return Storage.list(Storage.path("project"), "*.json")
    }
  }

  // Cache storage
  export namespace Cache {
    export function file(key: string): string {
      return Storage.path("cache", `${key}.json`)
    }
  }
}

13.3 Migration Pattern

export namespace Storage {
  type Migration = (dir: string) => Promise<void>

  const MIGRATIONS: Migration[] = [
    // Migration 1: Rename old directories
    async (dir) => {
      const oldPath = path.join(dir, "old-structure")
      const newPath = path.join(dir, "new-structure")

      if (await Filesystem.exists(oldPath)) {
        await fs.rename(oldPath, newPath)
        log.info("migrated directory structure")
      }
    },

    // Migration 2: Transform data format
    async (dir) => {
      const files = await Storage.list(path.join(dir, "sessions"), "*.json")

      for (const file of files) {
        const data = await Storage.read<any>(file)

        // Add new field
        if (!data.version) {
          data.version = 2
          data.migrated = Date.now()
          await Storage.write(file, data)
        }
      }

      log.info("migrated session data", { count: files.length })
    },

    // Migration 3: Consolidate files
    async (dir) => {
      const oldDir = path.join(dir, "old")
      const newFile = path.join(dir, "consolidated.json")

      if (await Filesystem.exists(oldDir)) {
        const files = await Storage.list(oldDir, "*.json")
        const consolidated = await Promise.all(files.map((f) => Storage.read(f)))

        await Storage.write(newFile, consolidated)
        await fs.rm(oldDir, { recursive: true })

        log.info("consolidated files", { count: files.length })
      }
    },
  ]

  export async function migrate(): Promise<void> {
    const dataDir = Global.Path.data
    const versionFile = path.join(dataDir, ".version")

    let currentVersion = 0
    if (await Filesystem.exists(versionFile)) {
      currentVersion = Number(await Bun.file(versionFile).text())
    }

    const targetVersion = MIGRATIONS.length

    if (currentVersion >= targetVersion) {
      log.debug("storage up to date", { version: currentVersion })
      return
    }

    log.info("migrating storage", {
      from: currentVersion,
      to: targetVersion,
    })

    // Run pending migrations
    for (let i = currentVersion; i < targetVersion; i++) {
      log.info(`running migration ${i + 1}/${targetVersion}`)
      await MIGRATIONS[i](dataDir)
    }

    // Update version
    await Bun.write(versionFile, String(targetVersion))

    log.info("migration complete", { version: targetVersion })
  }
}

File Reference: packages/opencode/src/storage/storage.ts:24-100

---

14. Error Handling Standards

14.1 Named Errors (Preferred)

// packages/opencode/src/util/error.ts

import { NamedError } from "@opencode-ai/util/error"
import z from "zod"

// Define typed errors
export const NotFoundError = NamedError.create(
  "NotFoundError",
  z.object({
    message: z.string(),
    resource: z.string().optional(),
    id: z.string().optional(),
  }),
)

export const ValidationError = NamedError.create(
  "ValidationError",
  z.object({
    message: z.string(),
    field: z.string().optional(),
    expected: z.string().optional(),
    received: z.string().optional(),
  }),
)

export const PermissionError = NamedError.create(
  "PermissionError",
  z.object({
    message: z.string(),
    action: z.string(),
    resource: z.string(),
  }),
)

// Usage: Throw typed error
throw NotFoundError.create({
  message: "Session not found",
  resource: "session",
  id: sessionID,
})

// Usage: Catch and check type
try {
  await operation()
} catch (error) {
  if (NotFoundError.is(error)) {
    log.warn("resource not found", {
      resource: error.data.resource,
      id: error.data.id,
    })
    return null
  }

  if (PermissionError.is(error)) {
    log.error("permission denied", {
      action: error.data.action,
      resource: error.data.resource,
    })
    throw error
  }

  // Unknown error
  log.error("unexpected error", { error })
  throw error
}

File Reference: packages/opencode/src/storage/storage.ts:17-22

14.2 Custom Error Classes (When Needed)

// ✅ GOOD - Error with additional context
export class BusyError extends Error {
  constructor(public readonly sessionID: string) {
    super(`Session ${sessionID} is busy`)
    this.name = "BusyError"
  }
}

// Usage
throw new BusyError(sessionID)

// Catching
try {
  await process(sessionID)
} catch (error) {
  if (error instanceof BusyError) {
    log.info("session busy, retrying", { sessionID: error.sessionID })
    await retry()
  }
}

File Reference: packages/opencode/src/session/index.ts:494

14.3 Result Pattern (for Tools)

// ✅ GOOD - Never throw in tool execution
export const ReadTool = Tool.define("read", {
  async execute(args, context) {
    try {
      const content = await Bun.file(args.filePath).text()

      return {
        output: content,
        title: `Read ${path.basename(args.filePath)}`,
      }
    } catch (error) {
      return {
        output: `Error reading file: ${error.message}`,
        title: "Read Failed",
        metadata: {
          error: true,
          message: error.message,
        },
      }
    }
  },
})

// ✅ GOOD - Result type for fallible operations
export type Result<T, E = Error> = { success: true; value: T } | { success: false; error: E }

export async function operation(): Promise<Result<Data>> {
  try {
    const data = await fetchData()
    return { success: true, value: data }
  } catch (error) {
    return { success: false, error }
  }
}

// Usage
const result = await operation()
if (result.success) {
  console.log(result.value)
} else {
  log.error("operation failed", { error: result.error })
}

14.4 Error Context

// ✅ GOOD - Rich error context
class OperationError extends Error {
  constructor(
    message: string,
    public readonly context: {
      operation: string
      input: unknown
      timestamp: number
      sessionID?: string
    }
  ) {
    super(message)
    this.name = "OperationError"
  }
}

// Usage
try {
  await processData(input)
} catch (error) {
  throw new OperationError(
    "Failed to process data",
    {
      operation: "processData",
      input,
      timestamp: Date.now(),
      sessionID: context.sessionID
    }
  )
}

// Logging with context
catch (error) {
  if (error instanceof OperationError) {
    log.error("operation failed", {
      operation: error.context.operation,
      sessionID: error.context.sessionID,
      error: error.message
    })
  }
}

---

15. Type System Standards

15.1 Zod Schemas (Primary)

Rule: Use Zod for runtime validation, derive TypeScript types from schemas

// ✅ GOOD - Schema-first approach
export namespace Session {
  export const Info = z.object({
    id: z.string(),
    title: z.string(),
    projectID: z.string(),
    modelID: z.string().optional(),
    permission: z.enum(["allow", "ask", "deny"]),
    time: z.object({
      created: z.number(),
      updated: z.number(),
    }),
  })

  // Derive TypeScript type from schema
  export type Info = z.infer<typeof Info>

  // Use in functions
  export const create = fn(Info.pick({ title: true, projectID: true }).partial(), async (input): Promise<Info> => {
    // Implementation
  })
}

// ❌ BAD - Type-first approach (no runtime validation)
export interface SessionInfo {
  id: string
  title: string
  projectID: string
}

export function create(input: Partial<SessionInfo>): SessionInfo {
  // No validation! Runtime errors possible
}

15.2 Type Inference

Rule: Rely on type inference, avoid explicit annotations

// ✅ GOOD - Let TypeScript infer
const session = await Session.create({ title: "New" })
const messages = session.messages.filter((m) => m.role === "user")
const ids = messages.map((m) => m.id)

// ❌ BAD - Unnecessary annotations
const session: Session.Info = await Session.create({ title: "New" })
const messages: Message[] = session.messages.filter((m) => m.role === "user")
const ids: string[] = messages.map((m) => m.id)

// ✅ ACCEPTABLE - Annotation for clarity in complex scenarios
const messages: Message[] = await fetchMessages().then((msgs): Message[] => {
  return msgs.filter((m) => m.deleted !== true)
})

File Reference: AGENTS.md:15

15.3 Avoid any

// ❌ BAD
function process(data: any) {
  return data.value.toString()
}

// ✅ GOOD - Use unknown and validate
function process(data: unknown) {
  if (typeof data !== "object" || data === null) {
    throw new Error("Expected object")
  }

  if (!("value" in data)) {
    throw new Error("Missing value property")
  }

  return String(data.value)
}

// ✅ BEST - Use Zod schema
const DataSchema = z.object({
  value: z.union([z.string(), z.number()]),
})

function process(data: unknown) {
  const validated = DataSchema.parse(data)
  return String(validated.value)
}

File Reference: AGENTS.md:12

15.4 Type Guards

// ✅ GOOD - Type guards for narrowing
function isMessage(value: unknown): value is Message {
  return typeof value === "object" && value !== null && "id" in value && "role" in value && typeof value.id === "string"
}

// Usage
if (isMessage(data)) {
  console.log(data.id) // TypeScript knows data is Message
}

// ✅ GOOD - Type guards in filter
const messages = items.filter((item): item is Message => isMessage(item)).map((msg) => msg.id)

// ✅ GOOD - Discriminated unions
type Part =
  | { type: "text"; text: string }
  | { type: "file"; data: Buffer; mimeType: string }
  | { type: "patch"; files: string[]; diff: string }

function handle(part: Part) {
  switch (part.type) {
    case "text":
      console.log(part.text) // TypeScript knows: text part
      break
    case "file":
      console.log(part.mimeType) // TypeScript knows: file part
      break
    case "patch":
      console.log(part.files) // TypeScript knows: patch part
      break
    default:
      const _exhaustive: never = part
      throw new Error("Unhandled part type")
  }
}

---

16. Import Organization Standards

16.1 Import Order

// 1. Zod and validation libraries
import z from "zod"

// 2. Node.js built-ins
import path from "path"
import fs from "fs/promises"
import os from "os"

// 3. External packages (alphabetical)
import { mergeDeep, sortBy, unique } from "remeda"
import chokidar from "chokidar"
import fuzzysort from "fuzzysort"

// 4. Internal packages (@opencode-ai/*)
import { NamedError } from "@opencode-ai/util/error"

// 5. Project utilities (../util/*)
import { Log } from "../util/log"
import { Lock } from "../util/lock"
import { fn } from "../util/fn"

// 6. Project services (../<service>/*)
import { Config } from "../config/config"
import { Storage } from "../storage/storage"
import { Bus } from "../bus"

// 7. Relative imports (same directory)
import { Session } from "./session"
import { MessageV2 } from "./message-v2"

16.2 Import Style

// ✅ GOOD - Named imports
import { create, update, get } from "./session"
import { Log, type LogOptions } from "./util/log"

// ❌ AVOID - Default imports (except for external packages)
import session from "./session"

// ✅ GOOD - Namespace imports for large APIs
import * as Session from "./session"

// ✅ GOOD - Type-only imports
import type { Session } from "./session"
import { type Config, loadConfig } from "./config"

16.3 Circular Dependency Prevention

// ❌ BAD - Circular dependency
// file-a.ts
import { funcB } from "./file-b"
export function funcA() {
  return funcB()
}

// file-b.ts
import { funcA } from "./file-a"
export function funcB() {
  return funcA()
}

// ✅ GOOD - Extract shared code
// shared.ts
export function shared() {
  return 42
}

// file-a.ts
import { shared } from "./shared"
export function funcA() {
  return shared()
}

// file-b.ts
import { shared } from "./shared"
export function funcB() {
  return shared()
}

---

17. Naming Conventions

17.1 Functions & Variables

// ✅ camelCase for functions and variables
const sessionID = "abc123"
const userAgent = "opencode/1.0"
const maxRetries = 3

async function processMessage(messageID: string) {}
async function createSession() {}

17.2 Types & Interfaces

// ✅ PascalCase for types, interfaces, and classes
type SessionInfo = {
  id: string
  title: string
}

interface Message {
  id: string
  content: string
}

class BusyError extends Error {}

namespace Session {
  export type Info = z.infer<typeof Info>
}

17.3 Constants

// ✅ SCREAMING_SNAKE_CASE for true constants
const MAX_RETRY_ATTEMPTS = 3
const DEFAULT_TIMEOUT = 5000
const API_BASE_URL = "https://api.example.com"

// ✅ camelCase for configuration values
const maxRetries = config.retries ?? 3
const defaultModel = config.model?.id ?? "gpt-4"

17.4 Namespaces

// ✅ PascalCase for namespaces
export namespace Session {}
export namespace Storage {}
export namespace FileWatcher {}

17.5 Files

// ✅ kebab-case for files
// session-manager.ts
// message-processor.ts
// file-watcher.ts

// ✅ Single word when possible
// session.ts
// storage.ts
// config.ts

---

18. Testing Standards

18.1 Principles

1. Avoid mocks - Test actual implementation
2. No logic duplication - Don't reimplement logic in tests
3. Use Bun test runner - bun test

File Reference: AGENTS.md:108-111

18.2 Test Structure

// packages/opencode/src/addons/serialize.test.ts

import { expect, test, describe } from "bun:test"
import { serialize, deserialize } from "./serialize"

describe("serialize", () => {
  test("preserves Date objects", () => {
    const input = { date: new Date("2024-01-01") }
    const serialized = serialize(input)
    const deserialized = deserialize(serialized)

    expect(deserialized.date).toBeInstanceOf(Date)
    expect(deserialized.date.toISOString()).toBe("2024-01-01T00:00:00.000Z")
  })

  test("preserves Set objects", () => {
    const input = { set: new Set([1, 2, 3]) }
    const serialized = serialize(input)
    const deserialized = deserialize(serialized)

    expect(deserialized.set).toBeInstanceOf(Set)
    expect(Array.from(deserialized.set)).toEqual([1, 2, 3])
  })

  test("handles nested structures", () => {
    const input = {
      map: new Map([["key", { date: new Date("2024-01-01") }]]),
      array: [new Set([1, 2])],
    }
    const serialized = serialize(input)
    const deserialized = deserialize(serialized)

    expect(deserialized.map).toBeInstanceOf(Map)
    expect(deserialized.map.get("key")?.date).toBeInstanceOf(Date)
    expect(deserialized.array[0]).toBeInstanceOf(Set)
  })
})

File Reference: packages/opencode/src/addons/serialize.test.ts

18.3 Test Commands

# Run all tests
bun test

# Run specific test file
bun test src/util/queue.test.ts

# Run tests matching pattern
bun test --test-name-pattern "serialize"

# Watch mode
bun test --watch

# Coverage
bun test --coverage

File Reference: packages/opencode/AGENTS.md - Build/Test Commands

18.4 Integration Tests

import { test, expect, beforeAll, afterAll } from "bun:test"
import { Session } from "./session"
import { Storage } from "./storage"

let sessionID: string

beforeAll(async () => {
  // Setup: Create test session
  const session = await Session.create({ title: "Test Session" })
  sessionID = session.id
})

afterAll(async () => {
  // Cleanup: Delete test session
  await Session.delete(sessionID)
})

test("create and retrieve session", async () => {
  const session = await Session.get(sessionID)

  expect(session).toBeDefined()
  expect(session.id).toBe(sessionID)
  expect(session.title).toBe("Test Session")
})

test("update session", async () => {
  await Session.update(sessionID, { title: "Updated" })

  const session = await Session.get(sessionID)
  expect(session.title).toBe("Updated")
})

---

19. Documentation Standards

19.1 Code Comments

Rule: Comment WHY, not WHAT

// ✅ GOOD - Explains why
// Cache header separately to enable 2-part prompt caching
if (system.length > 2 && system[0] === header) {
  system.length = 0
  system.push(header, rest.join("\n"))
}

// Prevent race condition: lock before checking existence
using _ = await Lock.write(filepath)
if (await Filesystem.exists(filepath)) {
  await fs.rm(filepath)
}

// ❌ BAD - States the obvious
// Push the item to the array
items.push(item)

// Set the title property
session.title = "New Title"

19.2 JSDoc (For Exported APIs)

/**
 * Execute a function with a file lock to prevent concurrent access.
 *
 * The lock is automatically released when the function completes or throws.
 * Locks are queued and processed in order.
 *
 * @param filepath - Absolute path to the file to lock
 * @param fn - Function to execute while holding the lock
 * @returns The result of the function
 *
 * @example
 * ```typescript
 * await withLock("/path/to/file", async () => {
 *   const content = await Bun.file("/path/to/file").text()
 *   await Bun.write("/path/to/file", content + "\n")
 * })
 * ```
 */
export async function withLock<T>(filepath: string, fn: () => Promise<T>): Promise<T>

19.3 README Structure

# Package Name

Brief description of what this package does.

## Installation

\`\`\`bash
bun install @opencode-ai/package-name
\`\`\`

## Usage

\`\`\`typescript
import { feature } from "@opencode-ai/package-name"

await feature()
\`\`\`

## API

### `function(param: Type): ReturnType`

Description of function.

**Parameters:**

- `param` - Description of parameter

**Returns:** Description of return value

**Example:**
\`\`\`typescript
const result = await function(param)
\`\`\`

## Development

\`\`\`bash
bun install
bun test
\`\`\`

---

20. Schema Definition Standards

20.1 Drizzle Schema (Database)

Rule: Use snake_case for field names to match SQL conventions

// ✅ GOOD - snake_case fields
const sessionTable = sqliteTable("session", {
  id: text().primaryKey(),
  project_id: text().notNull(),
  created_at: integer().notNull(),
  updated_at: integer().notNull(),
})

// ❌ BAD - camelCase requires explicit column names
const sessionTable = sqliteTable("session", {
  id: text("id").primaryKey(),
  projectID: text("project_id").notNull(),
  createdAt: integer("created_at").notNull(),
  updatedAt: integer("updated_at").notNull(),
})

File Reference: AGENTS.md:88-106

20.2 Zod Schema

// ✅ GOOD - Consistent field naming
export const SessionInfo = z.object({
  id: z.string(),
  projectID: z.string(), // camelCase for TypeScript
  title: z.string(),
  permission: z.enum(["allow", "ask", "deny"]),
  time: z.object({
    created: z.number(),
    updated: z.number(),
  }),
})

// ✅ GOOD - Schema composition
export const CreateSessionInput = SessionInfo.pick({
  title: true,
  projectID: true,
}).partial()

export const UpdateSessionInput = SessionInfo.partial().required({
  id: true,
})

// ✅ GOOD - Schema extension
export const SessionWithMessages = SessionInfo.extend({
  messages: z.array(MessageInfo),
})

---

21. Dependency Management Standards

21.1 Bun Runtime Preference

Rule: Use Bun APIs when available

// ✅ GOOD - Bun APIs
const content = await Bun.file(filepath).text()
await Bun.write(filepath, content)
const json = await Bun.file(filepath).json()

const proc = Bun.spawn(["ls", "-la"], {
  cwd: directory,
  stdout: "pipe",
})

// ❌ AVOID - Node.js fs when Bun alternative exists
import fs from "fs/promises"
const content = await fs.readFile(filepath, "utf-8")
await fs.writeFile(filepath, content)

File Reference: AGENTS.md:14

21.2 Package Installation

# Add dependency
bun add package-name

# Add dev dependency
bun add -d package-name

# Add workspace dependency
bun add @opencode-ai/other-package

21.3 Version Management

// package.json
{
  "dependencies": {
    "zod": "^3.22.4", // Allow patch and minor updates
    "remeda": "1.30.0", // Pin exact version for critical deps
    "@ai-sdk/anthropic": "^0.0.39"
  },
  "devDependencies": {
    "@types/node": "^20.10.0",
    "typescript": "^5.3.3"
  }
}

---

22. Build & Development Standards

22.1 Development Commands

# Install dependencies
bun install

# Run in development (TUI)
bun dev                       # Run in packages/opencode
bun dev <directory>           # Run in specific directory
bun dev .                     # Run in repo root

# Run API server
bun dev serve                 # Start on port 4096
bun dev serve --port 8080     # Custom port

# Run web UI (requires server running)
bun dev web                   # Start server + open web UI
bun run --cwd packages/app dev  # Just web UI

# Run desktop app
bun run --cwd packages/desktop tauri dev

# Type checking
bun run typecheck

# Run tests
bun test
bun test path/to/test.ts

# Build standalone executable
./packages/opencode/script/build.ts --single

# Regenerate SDK (after server changes)
./script/generate.ts

File Reference: CONTRIBUTING.md:30-150

22.2 Project Structure

opencode/
├── packages/
│   ├── opencode/           # Core business logic & CLI
│   │   ├── src/
│   │   │   ├── session/    # Session management
│   │   │   ├── tool/       # Tool implementations
│   │   │   ├── agent/      # Agent system
│   │   │   ├── mcp/        # Model Context Protocol
│   │   │   ├── lsp/        # Language Server Protocol
│   │   │   ├── file/       # File operations
│   │   │   ├── config/     # Configuration
│   │   │   ├── provider/   # LLM providers
│   │   │   ├── server/     # HTTP/WebSocket server
│   │   │   └── cli/cmd/tui/  # Terminal UI (SolidJS)
│   │   └── script/
│   │       └── build.ts    # Build script
│   ├── app/                # Web UI (SolidJS)
│   ├── desktop/            # Native app (Tauri)
│   ├── sdk/js/             # TypeScript SDK
│   ├── plugin/             # Plugin system
│   └── console/            # Web console
├── script/
│   ├── generate.ts         # Generate SDK
│   └── format.ts           # Format code
├── AGENTS.md               # Agent guidelines
├── CONTRIBUTING.md         # Contribution guide
└── CODEBASE_STANDARDS.md   # This document

22.3 Build Output

# Single build creates:
packages/opencode/dist/opencode-darwin-arm64/
├── bin/
│   └── opencode           # Executable
├── node_modules/          # Bundled dependencies
└── package.json

# Run built executable:
./packages/opencode/dist/opencode-darwin-arm64/bin/opencode

---

23. Performance Standards

23.1 Lazy Initialization

// ✅ GOOD - Lazy state initialization
const state = Instance.state(async () => {
  // Only initialize when first accessed
  const config = await Config.get()
  return { config, clients: [] }
})

// Access only when needed
if (needsClient) {
  const client = state().clients[0]
}

// ✅ GOOD - Lazy loading of heavy dependencies
const lazyModule = lazy(async () => {
  return import("./heavy-module")
})

// Load only when needed
const module = await lazyModule()

23.2 Caching

// ✅ GOOD - Cache expensive operations
const cache = new Map<string, Data>()

async function getData(id: string): Promise<Data> {
  const cached = cache.get(id)
  if (cached) return cached

  const data = await fetchData(id)
  cache.set(id, data)
  return data
}

// ✅ GOOD - Time-based cache invalidation
const cache = new Map<string, { data: Data; expires: number }>()

async function getData(id: string): Promise<Data> {
  const cached = cache.get(id)
  if (cached && Date.now() < cached.expires) {
    return cached.data
  }

  const data = await fetchData(id)
  cache.set(id, {
    data,
    expires: Date.now() + 60_000, // 1 minute
  })
  return data
}

23.3 Debouncing

// ✅ GOOD - Debounce high-frequency events
function debounce<T extends (...args: any[]) => any>(fn: T, delay: number): (...args: Parameters<T>) => void {
  let timer: Timer | undefined

  return (...args: Parameters<T>) => {
    if (timer) clearTimeout(timer)
    timer = setTimeout(() => {
      fn(...args)
      timer = undefined
    }, delay)
  }
}

// Usage
const debouncedSave = debounce(async (content: string) => {
  await save(content)
}, 500)

// Only saves after 500ms of inactivity
input.on("change", (content) => {
  debouncedSave(content)
})

23.4 Streaming

// ✅ GOOD - Stream large responses
export async function* streamMessages(sessionID: string) {
  const files = await Storage.list(Storage.path("messages", sessionID), "*.json")

  for (const file of files) {
    const message = await Storage.read<Message>(file)
    yield message
  }
}

// Usage
for await (const message of streamMessages(sessionID)) {
  process(message)
}

---

24. Security Standards

24.1 Input Validation

// ✅ GOOD - Validate all external input
export const handleRequest = fn(
  z.object({
    sessionID: z.string().regex(/^session-[a-z0-9]{10}$/),
    content: z.string().max(100_000),
    metadata: z.record(z.unknown()).optional(),
  }),
  async (input) => {
    // Input is validated, safe to use
    return process(input)
  },
)

// ✅ GOOD - Validate file paths
function validatePath(filepath: string) {
  const normalized = path.normalize(filepath)
  const absolute = path.resolve(normalized)

  // Prevent directory traversal
  if (!absolute.startsWith(Instance.worktree)) {
    throw new Error("Path outside worktree")
  }

  return absolute
}

24.2 Permission Checks

// ✅ GOOD - Check permissions before sensitive operations
async function readFile(filepath: string, context: Context) {
  // Validate path
  const safe = validatePath(filepath)

  // Check permission
  const permission = await PermissionNext.evaluate("read", safe, context.agent.permission)

  if (permission.action === "deny") {
    throw PermissionError.create({
      message: "Permission denied",
      action: "read",
      resource: safe,
    })
  }

  if (permission.action === "ask") {
    const allowed = await context.ask({
      type: "permission",
      message: `Read file ${path.basename(safe)}?`,
      actions: ["allow", "deny"],
    })

    if (!allowed) {
      throw PermissionError.create({
        message: "User denied permission",
        action: "read",
        resource: safe,
      })
    }
  }

  // Permission granted, perform operation
  return Bun.file(safe).text()
}

24.3 Secrets Management

// ✅ GOOD - Load secrets from environment
export namespace Auth {
  export async function get(providerID: string): Promise<string | undefined> {
    const key = `OPENCODE_${providerID.toUpperCase()}_API_KEY`
    return process.env[key]
  }
}

// ❌ AVOID - Hardcoded secrets
const API_KEY = "sk-abc123..."

// ❌ AVOID - Logging secrets
log.info("api key", { key: apiKey })

// ✅ GOOD - Redact secrets in logs
log.info("api request", { key: apiKey.slice(0, 8) + "..." })

24.4 Command Injection Prevention

// ❌ BAD - Command injection vulnerability
const output = await $`ls ${userInput}`.text()

// ✅ GOOD - Use array syntax (prevents injection)
const proc = Bun.spawn(["ls", userInput], {
  stdout: "pipe",
})
const output = await new Response(proc.stdout).text()

// ✅ GOOD - Validate input
const sanitized = userInput.replace(/[^a-zA-Z0-9_-]/g, "")
const proc = Bun.spawn(["ls", sanitized])

---

Summary: Key Principles

1. Single-word function names unless multi-word necessary (95%+ adherence)
2. Namespaces over classes for code organization (only 5 classes in 206 files)
3. Functional array methods over for-loops (85/15 split)
4. Parallel execution by default via Promise.all
5. Lock mechanisms prevent race conditions (reader-writer locks, file locks)
6. Per-instance state isolation for multi-project support
7. Event-driven architecture via pub/sub bus
8. Zod validation for all inputs via fn() wrapper
9. Disposable resources with using keyword for automatic cleanup
10. Bun runtime preference for file I/O and process spawning
11. Inline values once, extract when reusable
12. Named errors over throw strings
13. Avoid mocks in tests, test actual implementation
14. Layered config with clear precedence (7 levels)
15. Graceful disposal for all stateful services
16. Prefer const over let - use ternaries instead of reassignment
17. Avoid else statements - use early returns
18. Avoid destructuring - preserve context with dot notation
19. Type inference - avoid explicit annotations unless necessary
20. Avoid any type - use unknown with validation
21. Comment WHY, not WHAT - explain reasoning, not obvious operations
22. Schema-first approach - define Zod schemas, derive TS types
23. Reader-writer locks for concurrent access
24. System prompt caching - maintain 2-part structure for efficiency
25. Context compaction at 75% token threshold

---

File References Quick Index

Core Patterns

• Function wrapper: packages/opencode/src/util/fn.ts
• State management: packages/opencode/src/project/state.ts
• Instance isolation: packages/opencode/src/project/instance.ts
• Event bus: packages/opencode/src/bus/index.ts
• Locks: packages/opencode/src/util/lock.ts
• Async queue: packages/opencode/src/util/queue.ts

Services

• Session: packages/opencode/src/session/index.ts
• LLM streaming: packages/opencode/src/session/llm.ts
• Processor: packages/opencode/src/session/processor.ts
• Messages: packages/opencode/src/session/message-v2.ts
• LSP: packages/opencode/src/lsp/index.ts
• MCP: packages/opencode/src/mcp/index.ts
• Storage: packages/opencode/src/storage/storage.ts
• Config: packages/opencode/src/config/config.ts
• Provider: packages/opencode/src/provider/provider.ts

Tools

• Tool definition: packages/opencode/src/tool/tool.ts
• Edit: packages/opencode/src/tool/edit.ts
• Batch: packages/opencode/src/tool/batch.ts
• Grep: packages/opencode/src/tool/grep.ts

Documentation

• Style guide: AGENTS.md
• Contributing: CONTRIBUTING.md
• This document: CODEBASE_STANDARDS.md

---

End of Standards Document