feat: add delega init onboarding command#6
Conversation
Greptile SummaryThis PR adds Key observations:
Confidence Score: 3/5
Important Files Changed
Sequence DiagramsequenceDiagram
participant User
participant CLI as delega init
participant HostedAPI as api.delega.dev
participant Docker
participant LocalAPI as localhost:PORT
User->>CLI: npx delega init
alt Hosted path
CLI->>User: Prompt email
User->>CLI: email@example.com
CLI->>HostedAPI: POST /signup { email, name }
HostedAPI-->>CLI: 201 + agent.api_key
CLI->>User: Prompt 6-digit code (up to 3 attempts)
User->>CLI: 123456
CLI->>HostedAPI: POST /verify { email, code }
HostedAPI-->>CLI: 200 OK
CLI->>HostedAPI: POST /v1/tasks (demo task)
HostedAPI-->>CLI: 201 task
CLI->>CLI: persistApiKey() + saveConfig()
CLI->>User: Print API key box + MCP config
else Self-hosted path
CLI->>CLI: ensureDockerComposeInstalled()
CLI->>User: Prompt port [18890]
User->>CLI: port number
CLI->>CLI: writeComposeFile(port)
CLI->>Docker: docker compose up -d
Docker-->>CLI: containers starting (image pull if needed)
loop Poll /health (90s timeout, exp backoff)
CLI->>LocalAPI: GET /health
LocalAPI-->>CLI: 200 OK
end
CLI->>LocalAPI: POST /api/agents { name: "my-agent" }
LocalAPI-->>CLI: 201 + api_key
CLI->>LocalAPI: POST /api/tasks (demo task)
LocalAPI-->>CLI: 201 task
CLI->>CLI: persistApiKey() + saveConfig()
CLI->>User: Print API key box + MCP config
end
Last reviewed commit: "fix: Greptile round ..." |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
- waitForHealthy: 5×2s (~8s) → exponential backoff up to 90s with progress indicator. First-time Docker pulls can take 30-120s. - Remove unused promptSecret function (login.ts has its own copy).
|
@greptile please re-review — fixed both P1 (health check timeout bumped to 90s with exponential backoff) and P2 (removed dead promptSecret function). |
…F handling
- P0: Include DELEGA_API_URL in MCP config when self-hosted so
users don't get a broken config pointing at the hosted API
- P2: Add AbortSignal.timeout(15s) to requestJson and health-check
fetch to prevent indefinite hangs on unresponsive servers
- P2: Handle EOF (Ctrl+D) in promptText via rl.on('close') to
reject with InitCancelledError instead of hanging
|
@greptile re-review — commit 3d70a44 fixes all 5 issues from your previous review. Please review the latest commit. |
…sive template copy
- OTP verification: retry up to 3 attempts on wrong code instead of
locking user out (account exists server-side, reinit gets 409)
- parsePort: reject non-digit input like '8080x' that parseInt silently
truncates; use regex + Number() for strict validation
- copy-templates.mjs: use fs.cpSync({recursive: true}) to handle
subdirectories instead of silently skipping them
|
@greptile re-review — commit 329fb19 fixes all 3 remaining issues: OTP retry loop, strict port validation, recursive template copy. |
- P1: Move createDemoTask BEFORE storeApiKey/saveApiConfig so nothing is persisted to disk until the API confirms connectivity. On failure, the API key is printed to stderr so the user can recover manually. - Self-reviewed: walked every failure path in init.ts. Network errors, timeouts, SIGINT, EOF, disk write failures — all leave recoverable state. No orphaned config, no lost keys.
|
@greptile re-review — commit 9544250 fixes P1 credential persistence ordering. Self-reviewed all failure paths. |
| console.log(chalk.dim("Creating your first admin agent...")); | ||
|
|
||
| const agent = await bootstrapLocalAgent(apiBaseUrl); | ||
| return finalizeSetup(rawApiUrl, agent.api_key as string, rawApiUrl); |
There was a problem hiding this comment.
finalizeSetup is called with rawApiUrl for both the apiUrl and dashboardUrl arguments (e.g. http://localhost:18890). This URL is then surfaced to the user in printSuccess as:
Dashboard: http://localhost:18890
But http://localhost:18890 is the API root, not a web dashboard. Opening that URL in a browser will return API JSON or a 404, which is confusing immediately after a successful setup.
Compare with the hosted path where dashboardUrl = HOSTED_DASHBOARD_URL = "https://delega.dev/dashboard", which is a real, specific path.
If the Docker container exposes a dashboard at a known sub-path (e.g. /dashboard), pass that path here. If there is no web UI in the self-hosted image, omit the dashboard line for self-hosted instead of showing a non-functional link:
| return finalizeSetup(rawApiUrl, agent.api_key as string, rawApiUrl); | |
| return finalizeSetup(rawApiUrl, agent.api_key as string, `${rawApiUrl}/dashboard`); |
- P1: Check `docker compose version` instead of `docker --version` so Linux users without Compose plugin get a correct error message - P2: Port prompt now retries on invalid input instead of aborting - P2: Dashboard URL omitted for self-hosted (no web UI in Docker image) - P2: Docker image pinned to v1.0.0 via DELEGA_DOCKER_TAG constant (template uses __DELEGA_VERSION__ placeholder)
|
@greptile re-review — commit 189c144 fixes all 4 issues: Compose plugin check, port retry loop, no dashboard URL for self-hosted, pinned Docker image tag. Full adversarial self-review completed. |
…or handling - P1: Bind Docker port to 127.0.0.1 only — prevents API exposure on all network interfaces - P2: copy-templates.mjs wraps cpSync in try/catch with descriptive error and process.exit(1)
|
@greptile re-review — commit bee195a. Docker port bound to 127.0.0.1, build script error handling added. |
| ): Promise<{ response: Response; data: T | ApiErrorResponse }> { | ||
| let response: Response; | ||
| try { | ||
| response = await fetch(url, { ...init, signal: AbortSignal.timeout(FETCH_TIMEOUT_MS) }); |
There was a problem hiding this comment.
signal from caller is silently dropped
signal: AbortSignal.timeout(FETCH_TIMEOUT_MS) is placed after the ...init spread, so it unconditionally overwrites any signal property a caller passes in init. Currently no callers supply a signal, so there is no bug today — but if someone later needs to propagate a cancellation signal (e.g. an AbortController wired to Ctrl+C), it will be silently ignored and only the timeout will take effect.
Consider either combining the signals or documenting that caller-provided signals are not respected:
const callerSignal = (init as RequestInit & { signal?: AbortSignal }).signal;
const timeoutSignal = AbortSignal.timeout(FETCH_TIMEOUT_MS);
const signal = callerSignal
? AbortSignal.any([callerSignal, timeoutSignal])
: timeoutSignal;
response = await fetch(url, { ...init, signal });(AbortSignal.any is available in Node ≥ 20 — within the ">=18.0.0" range for projects that have already upgraded.)
| } from "../config.js"; | ||
| import { printBanner } from "../ui.js"; | ||
|
|
||
| const DELEGA_DOCKER_TAG = "1.0.0"; |
There was a problem hiding this comment.
DELEGA_DOCKER_TAG risks version drift
DELEGA_DOCKER_TAG is a hardcoded string "1.0.0" with no tie to the CLI's own package.json version. When a future CLI release adds API fields or changes behaviour that requires a matching server update, both this constant and package.json must be bumped together in the same commit — but there is no build-time check to enforce that.
Consider reading the intended server version from package.json (or a dedicated field like "engineVersion") at build time and embedding it as a generated constant, so the two are always kept in sync automatically.
| const email = await promptText("Your email: "); | ||
| if (!email) { | ||
| throw new UserFacingError("Email is required."); | ||
| } |
There was a problem hiding this comment.
Only emptiness is checked before sending the email to the API. An obviously malformed string like "not-an-email" or "@" passes through and produces whatever error message the server returns — which may be a raw validation message rather than a user-friendly hint.
Adding a minimal regex guard before the API call avoids a round-trip and gives the user a clear prompt to retry:
| const email = await promptText("Your email: "); | |
| if (!email) { | |
| throw new UserFacingError("Email is required."); | |
| } | |
| const email = await promptText("Your email: "); | |
| if (!email) { | |
| throw new UserFacingError("Email is required."); | |
| } | |
| if (!/^[^\s@]+@[^\s@]+\.[^\s@]+$/.test(email)) { | |
| throw new UserFacingError("Please enter a valid email address."); | |
| } |
| while (Date.now() - startTime < maxWaitMs) { | ||
| attempt += 1; | ||
| try { | ||
| const response = await fetch(healthUrl, { signal: AbortSignal.timeout(FETCH_TIMEOUT_MS) }); | ||
| if (response.ok) { | ||
| return; | ||
| } | ||
| } catch { | ||
| // Keep polling until the service responds or we run out of time. | ||
| } |
There was a problem hiding this comment.
maxWaitMs by up to FETCH_TIMEOUT_MS
The while condition guards loop entry (Date.now() - startTime < maxWaitMs), but the fetch call that immediately follows can itself block for up to FETCH_TIMEOUT_MS (15 s) beyond that deadline. In the worst case the function waits 90 s + 15 s = 105 s before throwing.
This is minor in practice, but the error message claims a 90-second timeout, which is misleading. The simplest fix is to check remaining time before issuing the fetch and break early if not enough time is left:
while (Date.now() - startTime < maxWaitMs) {
attempt += 1;
const remaining = maxWaitMs - (Date.now() - startTime);
try {
const response = await fetch(healthUrl, {
signal: AbortSignal.timeout(Math.min(FETCH_TIMEOUT_MS, remaining)),
});
if (response.ok) {
return;
}
} catch {
// Keep polling until the service responds or we run out of time.
}…ccess Default to 0.0.0.0 so other machines can reach the API. Added comment showing how to restrict to localhost if desired.
1 participant
Interactive onboarding flow for npx delega init. Supports hosted (api.delega.dev) and self-hosted (Docker) paths. 30-second time-to-value: email signup → verify → agent created → demo task → MCP config. Closes #394.