pr_01m47d24b0e6n91zwymwxg0vpx/services/runner/src/index.ts
| 1 | import { Container, type StopParams } from "@cloudflare/containers"; |
| 2 | import { WorkerEntrypoint } from "cloudflare:workers"; |
| 3 | |
| 4 | import { |
| 5 | type AgentModel, |
| 6 | type CheckJob, |
| 7 | type G1tEvent, |
| 8 | type Issue, |
| 9 | type Pull, |
| 10 | type RepoPath, |
| 11 | type Result, |
| 12 | type RunHostedInput, |
| 13 | type RunnerApi, |
| 14 | type ServiceBinding, |
| 15 | type User, |
| 16 | type Viewer, |
| 17 | fail, |
| 18 | identityClient, |
| 19 | ok, |
| 20 | workClient, |
| 21 | } from "@g1t/contracts"; |
| 22 | |
| 23 | import { type ConfiguredModel, modelEnv } from "./model-env"; |
| 24 | |
| 25 | export interface RunnerEnv { |
| 26 | SANDBOX: DurableObjectNamespace<AttemptSandbox>; |
| 27 | IDENTITY: ServiceBinding; |
| 28 | WORK: ServiceBinding; |
| 29 | /** Secret. The model key the hosted agent runs on. */ |
| 30 | ANTHROPIC_API_KEY?: string; |
| 31 | /** |
| 32 | * Comma-separated usernames allowed to start hosted agents. Runs spend the |
| 33 | * key above, so this stays an allowlist until accounts bring their own. |
| 34 | */ |
| 35 | HOSTED_AGENT_USERS: string; |
| 36 | /** |
| 37 | * The models offered, as JSON: |
| 38 | * `[{ id, label, description, modelName, model }]`. `modelName` is what |
| 39 | * people see; `model` is the identifier sent to the provider. |
| 40 | */ |
| 41 | AGENT_MODELS: string; |
| 42 | /** |
| 43 | * A Cloudflare AI Gateway id. When set, model traffic goes through that |
| 44 | * gateway, which is where logging, spend limits, caching and fallback |
| 45 | * between providers are configured. Empty sends it to the provider |
| 46 | * directly. |
| 47 | */ |
| 48 | AI_GATEWAY_ID: string; |
| 49 | CLOUDFLARE_ACCOUNT_ID: string; |
| 50 | /** Secret. Needed only if the gateway requires authentication. */ |
| 51 | AI_GATEWAY_TOKEN?: string; |
| 52 | } |
| 53 | |
| 54 | const MAX_AGENTS_PER_RUN = 5; |
| 55 | /** A run that takes longer than this has its token expire under it. */ |
| 56 | const TOKEN_TTL_SECONDS = 2 * 60 * 60; |
| 57 | /** How g1t's own agent is labelled. What runs behind it is g1t's choice. */ |
| 58 | const AGENT = "g1t-agent"; |
| 59 | |
| 60 | /** |
| 61 | * What a sandbox is doing: an agent working on a pull request as someone, |
| 62 | * or a run of acceptance checks. |
| 63 | */ |
| 64 | type Run = |
| 65 | | { kind: "agent"; actor: User; repo: RepoPath; number: number } |
| 66 | | { kind: "checks"; runId: string; token: string }; |
| 67 | type RunRequest = Run & { envVars: Record<string, string> }; |
| 68 | |
| 69 | /** Long enough to clone, install and test; then the token stops working. */ |
| 70 | const CHECKS_TOKEN_TTL_SECONDS = 45 * 60; |
| 71 | |
| 72 | /** |
| 73 | * One sandbox, for one agent or one run of checks. The image's entrypoint |
| 74 | * is the g1t runner, which does the work and exits; this class only starts |
| 75 | * it and cleans up if it dies without reporting. |
| 76 | */ |
| 77 | export class AttemptSandbox extends Container<RunnerEnv> { |
| 78 | sleepAfter = "45m"; |
| 79 | |
| 80 | async run(request: RunRequest): Promise<void> { |
| 81 | const { envVars, ...run } = request; |
| 82 | await this.ctx.storage.put("run", run); |
| 83 | await this.start({ envVars, enableInternet: true }); |
| 84 | } |
| 85 | |
| 86 | override async onStop({ exitCode }: StopParams): Promise<void> { |
| 87 | if (exitCode === 0) return; |
| 88 | const run = await this.ctx.storage.get<Run>("run"); |
| 89 | if (!run) return; |
| 90 | const work = workClient(this.env.WORK); |
| 91 | if (run.kind === "checks") { |
| 92 | // Refused harmlessly if the run did report before it stopped. |
| 93 | await work.reportChecks(run.runId, run.token, { |
| 94 | error: "The sandbox stopped before the checks finished.", |
| 95 | }); |
| 96 | return; |
| 97 | } |
| 98 | // The runner closes its own pull request when it fails. This covers a |
| 99 | // sandbox that was killed before it could; closing twice is refused |
| 100 | // harmlessly. |
| 101 | await work.closePull(run.actor, run.repo, run.number); |
| 102 | } |
| 103 | } |
| 104 | |
| 105 | function buildPrompt(issue: Issue, instructions: string): string { |
| 106 | const parts = [ |
| 107 | "You are a coding agent working in the git repository checked out in the current directory.", |
| 108 | `Issue #${issue.number}: ${issue.title}`, |
| 109 | issue.body, |
| 110 | ]; |
| 111 | if (issue.checks.length > 0) { |
| 112 | parts.push( |
| 113 | `These commands must pass when you are done. Run them if the tools are installed:\n${issue.checks.map((check) => `- ${check}`).join("\n")}`, |
| 114 | ); |
| 115 | } |
| 116 | if (instructions) parts.push(instructions); |
| 117 | parts.push( |
| 118 | "Make the change and keep it focused on the issue. Commit your work with a clear message. Do not push; that is done for you. Finish with a short summary of what you changed and why. It becomes the description of your pull request, so write it for a reviewer and leave out whether anything was committed or pushed.", |
| 119 | ); |
| 120 | return parts.filter(Boolean).join("\n\n"); |
| 121 | } |
| 122 | |
| 123 | export default class RunnerService |
| 124 | extends WorkerEntrypoint<RunnerEnv> |
| 125 | implements RunnerApi |
| 126 | { |
| 127 | /** A Worker must have an event handler; this service is RPC-only. */ |
| 128 | fetch(): Response { |
| 129 | return new Response("Not found\n", { status: 404 }); |
| 130 | } |
| 131 | |
| 132 | private configuredModels(): ConfiguredModel[] { |
| 133 | return JSON.parse(this.env.AGENT_MODELS); |
| 134 | } |
| 135 | |
| 136 | /** Whether sandboxes may be started on this person's say-so. */ |
| 137 | private enabledFor(username: string): boolean { |
| 138 | return this.env.HOSTED_AGENT_USERS.split(",") |
| 139 | .map((name) => name.trim()) |
| 140 | .includes(username); |
| 141 | } |
| 142 | |
| 143 | private allowed(viewer: Viewer): boolean { |
| 144 | if (!viewer || !this.env.ANTHROPIC_API_KEY) return false; |
| 145 | return this.enabledFor(viewer.username); |
| 146 | } |
| 147 | |
| 148 | /** Events from the bus: a pull request was opened, became ready, or moved. */ |
| 149 | async queue(batch: MessageBatch<G1tEvent>): Promise<void> { |
| 150 | for (const message of batch.messages) { |
| 151 | const event = message.body; |
| 152 | // A pull request opened from a branch is ready from the start; one |
| 153 | // opened as a draft is refused below until it is marked ready. |
| 154 | if ( |
| 155 | event.type === "pull.opened" || |
| 156 | event.type === "pull.ready" || |
| 157 | event.type === "pull.updated" |
| 158 | ) { |
| 159 | await this.startChecks(event.data.pullId); |
| 160 | } |
| 161 | message.ack(); |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | /** |
| 166 | * Runs a pull request's acceptance checks in a sandbox of its own. Does |
| 167 | * nothing when there is nothing to run. |
| 168 | */ |
| 169 | private async startChecks(pullId: string): Promise<boolean> { |
| 170 | const work = workClient(this.env.WORK); |
| 171 | const started = await work.startChecks(pullId); |
| 172 | if (!started.ok) return false; |
| 173 | const job: CheckJob = started.value; |
| 174 | // Checks are commands one person wrote, run against code another |
| 175 | // pushed, on g1t's machines. In the preview they run only when one of |
| 176 | // the two is someone sandboxes are enabled for. |
| 177 | if (!this.enabledFor(job.requestedBy) && !this.enabledFor(job.author.username)) { |
| 178 | await work.reportChecks(job.runId, job.token, { skip: true }); |
| 179 | return false; |
| 180 | } |
| 181 | // To read the commit, which may be private, as the one who pushed it. |
| 182 | const { token } = await identityClient(this.env.IDENTITY).createAccessToken( |
| 183 | job.author, |
| 184 | `Checks on ${job.repo.namespace}/${job.repo.name}#${job.number}`, |
| 185 | CHECKS_TOKEN_TTL_SECONDS, |
| 186 | ); |
| 187 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(job.runId)); |
| 188 | await sandbox.run({ |
| 189 | kind: "checks", |
| 190 | runId: job.runId, |
| 191 | token: job.token, |
| 192 | envVars: { |
| 193 | MODE: "checks", |
| 194 | G1T_API: "https://api.g1t.sh", |
| 195 | CHECK_RUN: job.runId, |
| 196 | CHECK_TOKEN: job.token, |
| 197 | G1T_USER: job.author.username, |
| 198 | G1T_TOKEN: token, |
| 199 | GIT_REMOTE: `https://g1t.sh/${job.source.namespace}/${job.source.name}.git`, |
| 200 | GIT_COMMIT: job.commit, |
| 201 | CHECKS: JSON.stringify(job.commands), |
| 202 | }, |
| 203 | }); |
| 204 | return true; |
| 205 | } |
| 206 | |
| 207 | async recheck(actor: User, repo: RepoPath, number: number): Promise<Result<boolean>> { |
| 208 | const found = await workClient(this.env.WORK).getPull(repo, number, actor); |
| 209 | if (!found.ok) return found; |
| 210 | const { pull } = found.value; |
| 211 | const member = (actor.workspaces ?? []).some( |
| 212 | (membership) => membership.slug === repo.namespace, |
| 213 | ); |
| 214 | if (!member && pull.author.id !== actor.id) { |
| 215 | return fail( |
| 216 | "forbidden", |
| 217 | "Only whoever opened a pull request, or a member of the workspace, can run its checks.", |
| 218 | ); |
| 219 | } |
| 220 | return (await this.startChecks(pull.id)) |
| 221 | ? ok(true) |
| 222 | : fail("conflict", "There are no checks to run for this pull request right now."); |
| 223 | } |
| 224 | |
| 225 | async models(viewer: Viewer): Promise<AgentModel[]> { |
| 226 | if (!this.allowed(viewer)) return []; |
| 227 | return this.configuredModels().map(({ id, label, description, modelName }) => ({ |
| 228 | id, |
| 229 | label, |
| 230 | description, |
| 231 | modelName, |
| 232 | })); |
| 233 | } |
| 234 | |
| 235 | async run( |
| 236 | actor: User, |
| 237 | repo: RepoPath, |
| 238 | issueNumber: number, |
| 239 | input: RunHostedInput, |
| 240 | ): Promise<Result<Pull[]>> { |
| 241 | if (!this.allowed(actor)) { |
| 242 | return fail("forbidden", "g1t agents are not enabled for your account."); |
| 243 | } |
| 244 | const models = this.configuredModels(); |
| 245 | const model = input.model |
| 246 | ? models.find((candidate) => candidate.id === input.model) |
| 247 | : models[0]; |
| 248 | if (!model) return fail("invalid", "That model is not available."); |
| 249 | const count = Math.min(Math.max(Math.trunc(input.count) || 1, 1), MAX_AGENTS_PER_RUN); |
| 250 | const identity = identityClient(this.env.IDENTITY); |
| 251 | const work = workClient(this.env.WORK); |
| 252 | |
| 253 | const found = await work.getIssue(repo, issueNumber, actor); |
| 254 | if (!found.ok) return found; |
| 255 | const { issue } = found.value; |
| 256 | const prompt = buildPrompt(issue, input.instructions?.trim() ?? ""); |
| 257 | |
| 258 | const pulls: Pull[] = []; |
| 259 | for (let i = 0; i < count; i++) { |
| 260 | const opened = await work.openPull(actor, repo, { |
| 261 | issue: issue.number, |
| 262 | agent: AGENT, |
| 263 | runtime: "hosted", |
| 264 | }); |
| 265 | // The first failure is the answer; later ones mean some already run. |
| 266 | if (!opened.ok) return pulls.length ? ok(pulls) : opened; |
| 267 | const pull = opened.value; |
| 268 | // Opened without a branch, so it has a fork. |
| 269 | const fork = pull.fork!; |
| 270 | pulls.push(pull); |
| 271 | |
| 272 | // The sandbox acts as the person who started it, through a token |
| 273 | // that only lives as long as a run can. |
| 274 | const { token } = await identity.createAccessToken( |
| 275 | actor, |
| 276 | `g1t agent on ${repo.namespace}/${repo.name}#${pull.number}`, |
| 277 | TOKEN_TTL_SECONDS, |
| 278 | ); |
| 279 | const sandbox = this.env.SANDBOX.get(this.env.SANDBOX.idFromName(pull.id)); |
| 280 | await sandbox.run({ |
| 281 | kind: "agent", |
| 282 | actor, |
| 283 | repo, |
| 284 | number: pull.number, |
| 285 | envVars: { |
| 286 | G1T_API: "https://api.g1t.sh", |
| 287 | G1T_TOKEN: token, |
| 288 | G1T_USER: actor.username, |
| 289 | G1T_REPO: `${repo.namespace}/${repo.name}`, |
| 290 | PULL_NUMBER: String(pull.number), |
| 291 | GIT_REMOTE: `https://g1t.sh/${fork.namespace}/${fork.name}.git`, |
| 292 | COMMIT_MESSAGE: issue.title, |
| 293 | PROMPT: prompt, |
| 294 | ...modelEnv(this.env, model), |
| 295 | }, |
| 296 | }); |
| 297 | } |
| 298 | return ok(pulls); |
| 299 | } |
| 300 | } |