feat(core): add scout agent for repo research (#24149)

Co-authored-by: Dax Raad <d@ironbay.co>
This commit is contained in:
Shoubhit Dash
2026-05-08 20:20:08 +00:00
committed by GitHub
co-authored by Dax Raad
parent 6e47ae769e
commit 40d5ea1cf1
44 changed files with 1622 additions and 66 deletions
+63
View File
@@ -0,0 +1,63 @@
import { Effect, Schema } from "effect"
import { HttpClient } from "effect/unstable/http"
import * as Tool from "./tool"
import * as McpWebSearch from "./mcp-websearch"
import DESCRIPTION from "./codesearch.txt"
export const Parameters = Schema.Struct({
query: Schema.String.annotate({
description:
"Search query to find relevant context for APIs, Libraries, and SDKs. For example, 'React useState hook examples', 'Python pandas dataframe filtering', 'Express.js middleware', 'Next js partial prerendering configuration'",
}),
tokensNum: Schema.Number.check(Schema.isGreaterThanOrEqualTo(1000))
.check(Schema.isLessThanOrEqualTo(50000))
.pipe(Schema.optional, Schema.withDecodingDefault(Effect.succeed(5000)))
.annotate({
description:
"Number of tokens to return (1000-50000). Default is 5000 tokens. Adjust this value based on how much context you need - use lower values for focused queries and higher values for comprehensive documentation.",
}),
})
export const CodeSearchTool = Tool.define(
"codesearch",
Effect.gen(function* () {
const http = yield* HttpClient.HttpClient
return {
description: DESCRIPTION,
parameters: Parameters,
execute: (params: { query: string; tokensNum: number }, ctx: Tool.Context) =>
Effect.gen(function* () {
yield* ctx.ask({
permission: "codesearch",
patterns: [params.query],
always: ["*"],
metadata: {
query: params.query,
tokensNum: params.tokensNum,
},
})
const result = yield* McpWebSearch.call(
http,
McpWebSearch.EXA_URL,
"get_code_context_exa",
McpWebSearch.CodeArgs,
{
query: params.query,
tokensNum: params.tokensNum,
},
"30 seconds",
)
return {
output:
result ??
"No code snippets or documentation found. Please try a different query, be more specific about the library or programming concept, or check the spelling of framework names.",
title: `Code search: ${params.query}`,
metadata: {},
}
}).pipe(Effect.orDie),
}
}),
)
+12
View File
@@ -0,0 +1,12 @@
- Search and get relevant context for any programming task using Exa Code API
- Provides the highest quality and freshest context for libraries, SDKs, and APIs
- Use this tool for ANY question or task related to programming
- Returns comprehensive code examples, documentation, and API references
- Optimized for finding specific programming patterns and solutions
Usage notes:
- Adjustable token count (1000-50000) for focused or comprehensive results
- Default 5000 tokens provides balanced context for most queries
- Use lower values for specific questions, higher values for comprehensive documentation
- Supports queries about frameworks, libraries, APIs, and programming concepts
- Examples: 'React useState hook examples', 'Python pandas dataframe filtering', 'Express.js middleware'
@@ -48,6 +48,11 @@ export const SearchArgs = Schema.Struct({
contextMaxCharacters: Schema.optional(Schema.Number),
})
export const CodeArgs = Schema.Struct({
query: Schema.String,
tokensNum: Schema.Number,
})
export const ParallelSearchArgs = Schema.Struct({
objective: Schema.String,
search_queries: Schema.Array(Schema.String),
+13
View File
@@ -22,6 +22,9 @@ import { Plugin } from "../plugin"
import { Provider } from "@/provider/provider"
import { ProviderID, type ModelID } from "../provider/schema"
import { WebSearchTool } from "./websearch"
import { CodeSearchTool } from "./codesearch"
import { RepoCloneTool } from "./repo_clone"
import { RepoOverviewTool } from "./repo_overview"
import { Flag } from "@opencode-ai/core/flag/flag"
import * as Log from "@opencode-ai/core/util/log"
import { LspTool } from "./lsp"
@@ -44,6 +47,7 @@ import { Instruction } from "../session/instruction"
import { AppFileSystem } from "@opencode-ai/core/filesystem"
import { Bus } from "../bus"
import { Agent } from "../agent/agent"
import { Git } from "@/git"
import { Skill } from "../skill"
import { Permission } from "@/permission"
@@ -86,6 +90,7 @@ export const layer: Layer.Layer<
| Skill.Service
| Session.Service
| Provider.Service
| Git.Service
| LSP.Service
| Instruction.Service
| AppFileSystem.Service
@@ -113,6 +118,9 @@ export const layer: Layer.Layer<
const plan = yield* PlanExitTool
const webfetch = yield* WebFetchTool
const websearch = yield* WebSearchTool
const codesearch = yield* CodeSearchTool
const repoClone = yield* RepoCloneTool
const repoOverview = yield* RepoOverviewTool
const shell = yield* ShellTool
const globtool = yield* GlobTool
const writetool = yield* WriteTool
@@ -212,6 +220,9 @@ export const layer: Layer.Layer<
fetch: Tool.init(webfetch),
todo: Tool.init(todo),
search: Tool.init(websearch),
code: Tool.init(codesearch),
repo_clone: Tool.init(repoClone),
repo_overview: Tool.init(repoOverview),
skill: Tool.init(skilltool),
patch: Tool.init(patchtool),
question: Tool.init(question),
@@ -234,6 +245,7 @@ export const layer: Layer.Layer<
tool.fetch,
tool.todo,
tool.search,
...(Flag.OPENCODE_EXPERIMENTAL_SCOUT ? [tool.code, tool.repo_clone, tool.repo_overview] : []),
tool.skill,
tool.patch,
...(Flag.OPENCODE_EXPERIMENTAL_LSP_TOOL ? [tool.lsp] : []),
@@ -348,6 +360,7 @@ export const defaultLayer = Layer.suspend(() =>
Layer.provide(Agent.defaultLayer),
Layer.provide(Session.defaultLayer),
Layer.provide(Provider.defaultLayer),
Layer.provide(Git.defaultLayer),
Layer.provide(LSP.defaultLayer),
Layer.provide(Instruction.defaultLayer),
Layer.provide(AppFileSystem.defaultLayer),
+209
View File
@@ -0,0 +1,209 @@
import path from "path"
import { Effect, Schema } from "effect"
import { AppFileSystem } from "@opencode-ai/core/filesystem"
import { Flock } from "@opencode-ai/core/util/flock"
import { Git } from "@/git"
import DESCRIPTION from "./repo_clone.txt"
import * as Tool from "./tool"
import { parseRepositoryReference, repositoryCachePath, sameRepositoryReference } from "@/util/repository"
export const Parameters = Schema.Struct({
repository: Schema.String.annotate({
description: "Repository to clone, as a git URL, host/path reference, or GitHub owner/repo shorthand",
}),
refresh: Schema.optional(Schema.Boolean).annotate({
description: "When true, fetches the latest remote state into the managed cache",
}),
branch: Schema.optional(Schema.String).annotate({
description: "Branch or ref to clone and inspect",
}),
})
type Metadata = {
repository: string
host: string
remote: string
localPath: string
status: "cached" | "cloned" | "refreshed"
head?: string
branch?: string
}
function statusForRepository(input: { reuse: boolean; refresh?: boolean; branchMatches?: boolean }) {
if (!input.reuse) return "cloned" as const
if (input.branchMatches === false) return "refreshed" as const
if (input.refresh) return "refreshed" as const
return "cached" as const
}
function resetTarget(input: {
requestedBranch?: string
remoteHead: { code: number; stdout: string }
branch: { code: number; stdout: string }
}) {
if (input.requestedBranch) return `origin/${input.requestedBranch}`
if (input.remoteHead.code === 0 && input.remoteHead.stdout) {
return input.remoteHead.stdout.replace(/^refs\/remotes\//, "")
}
if (input.branch.code === 0 && input.branch.stdout) {
return `origin/${input.branch.stdout}`
}
return "HEAD"
}
function validateBranch(branch: string) {
if (!/^[A-Za-z0-9/_.-]+$/.test(branch) || branch.startsWith("-") || branch.includes("..")) {
throw new Error(
"Branch must contain only alphanumeric characters, /, _, ., and -, and cannot start with - or contain ..",
)
}
}
export const RepoCloneTool = Tool.define<typeof Parameters, Metadata, AppFileSystem.Service | Git.Service>(
"repo_clone",
Effect.gen(function* () {
const fs = yield* AppFileSystem.Service
const git = yield* Git.Service
return {
description: DESCRIPTION,
parameters: Parameters,
execute: (params: Schema.Schema.Type<typeof Parameters>, ctx: Tool.Context<Metadata>) =>
Effect.gen(function* () {
const reference = parseRepositoryReference(params.repository)
if (!reference)
throw new Error("Repository must be a git URL, host/path reference, or GitHub owner/repo shorthand")
if (reference.protocol === "file:") throw new Error("Local file repositories are not supported")
if (params.branch) validateBranch(params.branch)
const repository = reference.label
const remote = reference.remote
const localPath = repositoryCachePath(reference)
const cloneTarget = parseRepositoryReference(remote) ?? reference
yield* ctx.ask({
permission: "repo_clone",
patterns: [repository],
always: [repository],
metadata: {
repository,
remote,
path: localPath,
refresh: Boolean(params.refresh),
branch: params.branch,
},
})
return yield* Effect.acquireUseRelease(
Effect.promise((signal) => Flock.acquire(`repo-clone:${localPath}`, { signal })),
() =>
Effect.gen(function* () {
yield* fs.ensureDir(path.dirname(localPath)).pipe(Effect.orDie)
const exists = yield* fs.existsSafe(localPath)
const hasGitDir = yield* fs.existsSafe(path.join(localPath, ".git"))
const origin = hasGitDir
? yield* git.run(["config", "--get", "remote.origin.url"], { cwd: localPath })
: undefined
const originReference =
origin?.exitCode === 0 ? parseRepositoryReference(origin.text().trim()) : undefined
const reuse =
hasGitDir && Boolean(originReference && sameRepositoryReference(originReference, cloneTarget))
if (exists && !reuse) {
yield* fs.remove(localPath, { recursive: true }).pipe(Effect.orDie)
}
const currentBranch = hasGitDir ? yield* git.branch(localPath) : undefined
const status = statusForRepository({
reuse,
refresh: params.refresh,
branchMatches: params.branch ? currentBranch === params.branch : undefined,
})
if (status === "cloned") {
const clone = yield* git.run(
[
"clone",
"--depth",
"100",
...(params.branch ? ["--branch", params.branch] : []),
"--",
remote,
localPath,
],
{ cwd: path.dirname(localPath) },
)
if (clone.exitCode !== 0) {
throw new Error(
clone.stderr.toString().trim() || clone.text().trim() || `Failed to clone ${repository}`,
)
}
}
if (status === "refreshed") {
const fetch = yield* git.run(["fetch", "--all", "--prune"], { cwd: localPath })
if (fetch.exitCode !== 0) {
throw new Error(
fetch.stderr.toString().trim() || fetch.text().trim() || `Failed to refresh ${repository}`,
)
}
if (params.branch) {
const checkout = yield* git.run(["checkout", "-B", params.branch, `origin/${params.branch}`], {
cwd: localPath,
})
if (checkout.exitCode !== 0) {
throw new Error(
checkout.stderr.toString().trim() ||
checkout.text().trim() ||
`Failed to checkout ${params.branch}`,
)
}
}
const remoteHead = yield* git.run(["symbolic-ref", "refs/remotes/origin/HEAD"], { cwd: localPath })
const branch = yield* git.run(["symbolic-ref", "--quiet", "--short", "HEAD"], { cwd: localPath })
const target = resetTarget({
requestedBranch: params.branch,
remoteHead: { code: remoteHead.exitCode, stdout: remoteHead.text().trim() },
branch: { code: branch.exitCode, stdout: branch.text().trim() },
})
const reset = yield* git.run(["reset", "--hard", target], { cwd: localPath })
if (reset.exitCode !== 0) {
throw new Error(
reset.stderr.toString().trim() || reset.text().trim() || `Failed to reset ${repository}`,
)
}
}
const head = yield* git.run(["rev-parse", "HEAD"], { cwd: localPath })
const branch = yield* git.branch(localPath)
const headText = head.exitCode === 0 ? head.text().trim() : undefined
return {
title: repository,
metadata: {
repository,
host: reference.host,
remote,
localPath,
status,
head: headText,
branch,
},
output: [
`Repository ready: ${repository}`,
`Status: ${status}`,
`Local path: ${localPath}`,
...(branch ? [`Branch: ${branch}`] : []),
...(headText ? [`HEAD: ${headText}`] : []),
].join("\n"),
}
}),
(lock) => Effect.promise(() => lock.release()).pipe(Effect.ignore),
)
}).pipe(Effect.orDie),
} satisfies Tool.DefWithoutID<typeof Parameters, Metadata>
}),
)
@@ -0,0 +1,5 @@
- Clone or refresh a repository into OpenCode's managed cache under the data directory
- Accepts git URLs, forge host/path references, or GitHub owner/repo shorthand
- Returns the cached absolute local path so other tools can explore the cloned source
- Use this before Read, Glob, or Grep when the code you need lives outside the current workspace
- This tool is intended for dependency and documentation research workflows, not for modifying the user's workspace
+238
View File
@@ -0,0 +1,238 @@
import path from "path"
import { Effect, Schema } from "effect"
import { AppFileSystem } from "@opencode-ai/core/filesystem"
import { Git } from "@/git"
import { assertExternalDirectoryEffect } from "./external-directory"
import DESCRIPTION from "./repo_overview.txt"
import * as Tool from "./tool"
import { parseRepositoryReference, repositoryCachePath } from "@/util/repository"
import { Instance } from "@/project/instance"
export const Parameters = Schema.Struct({
repository: Schema.optional(Schema.String).annotate({
description: "Cached repository to inspect, as a git URL, host/path reference, or GitHub owner/repo shorthand",
}),
path: Schema.optional(Schema.String).annotate({
description: "Directory path to inspect instead of a cached repository",
}),
depth: Schema.optional(Schema.Number).annotate({
description: "Maximum structure depth to include. Defaults to 3.",
})
})
type Metadata = {
path: string
repository?: string
branch?: string
head?: string
package_manager?: string
ecosystems: string[]
dependency_files: string[]
entrypoints: string[]
depth: number
truncated: boolean
}
const IGNORED_DIRS = new Set([".git", "node_modules", "__pycache__", ".venv", "dist", "build", ".next", "target", "vendor"])
const STRUCTURE_LIMIT = 200
const DEPENDENCY_FILES = [
"package.json",
"package-lock.json",
"bun.lock",
"bun.lockb",
"pnpm-lock.yaml",
"yarn.lock",
"requirements.txt",
"pyproject.toml",
"go.mod",
"Cargo.toml",
"Gemfile",
"build.gradle",
"build.gradle.kts",
"pom.xml",
"composer.json",
]
function packageManager(files: Set<string>) {
if (files.has("bun.lock") || files.has("bun.lockb")) return "bun"
if (files.has("pnpm-lock.yaml")) return "pnpm"
if (files.has("yarn.lock")) return "yarn"
if (files.has("package-lock.json")) return "npm"
}
function ecosystems(files: Set<string>) {
return [
...(files.has("package.json") ? ["Node.js"] : []),
...(files.has("pyproject.toml") || files.has("requirements.txt") ? ["Python"] : []),
...(files.has("go.mod") ? ["Go"] : []),
...(files.has("Cargo.toml") ? ["Rust"] : []),
...(files.has("Gemfile") ? ["Ruby"] : []),
...(files.has("build.gradle") || files.has("build.gradle.kts") || files.has("pom.xml") ? ["Java/Kotlin"] : []),
...(files.has("composer.json") ? ["PHP"] : []),
]
}
function commonEntrypoints(files: Set<string>) {
return ["index.ts", "index.tsx", "index.js", "index.mjs", "main.ts", "main.js", "src/index.ts", "src/index.tsx", "src/index.js", "src/main.ts", "src/main.js"].filter((file) => files.has(file))
}
export const RepoOverviewTool = Tool.define<typeof Parameters, Metadata, AppFileSystem.Service | Git.Service>(
"repo_overview",
Effect.gen(function* () {
const fs = yield* AppFileSystem.Service
const git = yield* Git.Service
const resolveTarget = Effect.fn("RepoOverviewTool.resolveTarget")(function* (params: Schema.Schema.Type<typeof Parameters>) {
if (params.path) {
const full = path.isAbsolute(params.path) ? params.path : path.resolve(Instance.directory, params.path)
return { path: full, repository: params.repository }
}
if (!params.repository) throw new Error("Either repository or path is required")
const parsed = parseRepositoryReference(params.repository)
if (!parsed) throw new Error("Repository must be a git URL, host/path reference, or GitHub owner/repo shorthand")
const repository = parsed.label
return {
repository,
path: repositoryCachePath(parsed),
}
})
const structure = Effect.fn("RepoOverviewTool.structure")(function* (root: string, depth: number) {
let truncated = false
const lines: string[] = []
const visit: (dir: string, level: number) => Effect.Effect<void> = Effect.fnUntraced(function* (dir: string, level: number) {
if (level >= depth || lines.length >= STRUCTURE_LIMIT) {
truncated = truncated || lines.length >= STRUCTURE_LIMIT
return
}
const entries = yield* fs.readDirectoryEntries(dir).pipe(Effect.orElseSucceed(() => []))
const sorted = yield* Effect.forEach(
entries,
Effect.fnUntraced(function* (entry) {
if (IGNORED_DIRS.has(entry.name)) return undefined
const full = path.join(dir, entry.name)
const info = yield* fs.stat(full).pipe(Effect.catch(() => Effect.succeed(undefined)))
if (!info) return undefined
return { name: entry.name, full, directory: info.type === "Directory" }
}),
{ concurrency: 16 },
).pipe(
Effect.map((items) =>
items
.filter((item): item is { name: string; full: string; directory: boolean } => Boolean(item))
.sort((a, b) => Number(b.directory) - Number(a.directory) || a.name.localeCompare(b.name)),
),
)
for (const entry of sorted) {
if (lines.length >= STRUCTURE_LIMIT) {
truncated = true
return
}
lines.push(`${" ".repeat(level)}${entry.name}${entry.directory ? "/" : ""}`)
if (entry.directory) yield* visit(entry.full, level + 1)
}
})
yield* visit(root, 0)
return { lines, truncated }
})
return {
description: DESCRIPTION,
parameters: Parameters,
execute: (params: Schema.Schema.Type<typeof Parameters>, ctx: Tool.Context<Metadata>) =>
Effect.gen(function* () {
const target = yield* resolveTarget(params)
const depth = !params.depth || !Number.isInteger(params.depth) || params.depth < 1 || params.depth > 6 ? 3 : params.depth
yield* assertExternalDirectoryEffect(ctx, target.path, { kind: "directory" })
yield* ctx.ask({
permission: "repo_overview",
patterns: [target.repository ?? target.path],
always: [target.repository ?? target.path],
metadata: {
repository: target.repository,
path: target.path,
depth,
},
})
const info = yield* fs.stat(target.path).pipe(Effect.catch(() => Effect.succeed(undefined)))
if (!info) {
if (target.repository) throw new Error(`Repository is not cloned: ${target.repository}. Use repo_clone first.`)
throw new Error(`Directory not found: ${target.path}`)
}
if (info.type !== "Directory") throw new Error(`Path is not a directory: ${target.path}`)
const entries = yield* fs.readDirectoryEntries(target.path).pipe(Effect.orElseSucceed(() => []))
const topLevel = new Set(entries.map((entry) => entry.name))
const dependencyFiles = DEPENDENCY_FILES.filter((file) => topLevel.has(file))
const packageJson = topLevel.has("package.json")
? (yield* fs.readJson(path.join(target.path, "package.json")).pipe(Effect.orElseSucceed(() => ({})))) as Record<string, unknown>
: {}
const entrypoints = [
...(typeof packageJson.main === "string" ? [`main: ${packageJson.main}`] : []),
...(typeof packageJson.module === "string" ? [`module: ${packageJson.module}`] : []),
...(typeof packageJson.types === "string" ? [`types: ${packageJson.types}`] : []),
...(typeof packageJson.bin === "string" ? [`bin: ${packageJson.bin}`] : []),
...(packageJson.bin && typeof packageJson.bin === "object" && !Array.isArray(packageJson.bin)
? Object.keys(packageJson.bin as Record<string, unknown>).map((name) => `bin: ${name}`)
: []),
...(packageJson.exports && typeof packageJson.exports === "object" && !Array.isArray(packageJson.exports)
? Object.keys(packageJson.exports as Record<string, unknown>).slice(0, 10).map((name) => `exports: ${name}`)
: []),
]
const common = commonEntrypoints(new Set([
...topLevel,
...entries
.filter((entry) => entry.name === "src")
.flatMap(() => ["src/index.ts", "src/index.tsx", "src/index.js", "src/main.ts", "src/main.js"]),
]))
const structureResult = yield* structure(target.path, depth)
const branch = yield* git.branch(target.path)
const head = yield* git.run(["rev-parse", "HEAD"], { cwd: target.path })
const headText = head.exitCode === 0 ? head.text().trim() : undefined
const metadata: Metadata = {
path: target.path,
repository: target.repository,
branch,
head: headText,
package_manager: packageManager(topLevel),
ecosystems: ecosystems(topLevel),
dependency_files: dependencyFiles,
entrypoints: [...entrypoints, ...common.map((file) => `file: ${file}`)],
depth,
truncated: structureResult.truncated,
}
return {
title: target.repository ?? path.basename(target.path),
metadata,
output: [
`Path: ${target.path}`,
...(target.repository ? [`Repository: ${target.repository}`] : []),
...(branch ? [`Branch: ${branch}`] : []),
...(headText ? [`HEAD: ${headText}`] : []),
...(metadata.ecosystems.length ? [`Ecosystems: ${metadata.ecosystems.join(", ")}`] : []),
...(metadata.package_manager ? [`Package manager: ${metadata.package_manager}`] : []),
...(metadata.dependency_files.length ? [`Dependency files: ${metadata.dependency_files.join(", ")}`] : []),
...(metadata.entrypoints.length ? ["Likely entrypoints:", ...metadata.entrypoints.map((entry) => `- ${entry}`)] : []),
"Top-level structure:",
...structureResult.lines,
...(structureResult.truncated ? ["(Structure truncated)"] : []),
].join("\n"),
}
}).pipe(Effect.orDie),
} satisfies Tool.DefWithoutID<typeof Parameters, Metadata>
}),
)
@@ -0,0 +1,4 @@
- Summarize the structure and likely entrypoints of a cloned repository or local directory
- Accepts either a cached repository reference or a directory path
- Reports detected ecosystems, dependency files, package manager, likely entrypoints, and a compact structure tree
- Use this after repo_clone to orient quickly before deeper Read, Glob, or Grep investigation