> ## Documentation Index > Fetch the complete documentation index at: https://docs.pome.sh/llms.txt > Use this file to discover all available pages before exploring further. # GitHub twin in 5 minutes > Start a private GitHub, let your own coding agent triage the open bug in it, then read back the twin's recorded tape of every call it made. No GitHub account, no API key, nothing graded. **About 5 minutes.** You start a private GitHub holding one open bug report, your own coding agent triages it, and you read the twin's own record of every call it made — including the `201` on the comment it wrote and the state that changed behind it. A **digital twin** is not a mock. It is a stateful service that answers the same REST and MCP calls as `api.github.com`, boots from a declared starting state, records every request, and never touches github.com. What you write at one call is there at the next — and gone when the sandbox stops. ## Before you start * **A Pome account.** `pome login` opens the browser sign-in and creates one if you do not have it. No credit card. * **Your own coding agent** — Claude Code, Cursor, anything that can run a shell command and read JSON back. * **Node 18+** for `npx`, plus `curl` and `jq` for the transcripts below. Your agent can read the raw JSON without `jq`; it is here to keep the blocks short. Nothing else. No GitHub account, no `ANTHROPIC_API_KEY`, no model inference paid for by Pome: **your agent is both the operator and the actor.** It drives Pome, and it is the thing that acts on the twin. **Nothing on this page is graded.** No task file, no criteria, no score — and no agent eval is charged, because an eval is only ever burned when a run is graded. A sandbox you start, drive and stop costs you nothing. Grading appears exactly once in this curriculum, at the [support-triage capstone](/quickstart/claude-code), where the agent under test is sealed off from the criteria that judge it. ## Paste this Hand this to your coding agent as-is. It names the twin's own action vocabulary (`add_issue_comment`, `add_assignees`) and the boundary it must not cross. ```text theme={"dark"} Set up a Pome GitHub sandbox, triage the bug already in it, then show me the recorded tape. Show me this plan first and wait for my go-ahead. Keep every command plain text; never pipe a remote script into a shell. 1. Sign in and start the world: npx @pome-sh/cli@latest login npx @pome-sh/cli@latest sandbox create --twin github \ --secrets-file .pome-sandbox.env --format json Then source .pome-sandbox.env. POME_AUTH_TOKEN is the bearer for EVERY call to the twin and the only one it accepts: any other token, or none, comes back as an opaque 404. Keep it in the shell, and add .pome-sandbox.env to .gitignore. 2. Read the world before you touch it: GET $POME_GITHUB_REST_URL/repos/acme/api/issues Tell me what is open and how it is labelled. 3. Triage that issue the way you would triage a real one, through the twin's own GitHub surface -- the same routes and payloads api.github.com takes: add_issue_comment POST .../issues//comments, with a repro summary drawn from the issue body, not invented. add_assignees POST .../issues//assignees, ["alice"]. Do not open a new issue and do not close this one. 4. Read the result back THROUGH THE TWIN, not from your memory of what you sent: GET .../issues/ and .../issues//comments. 5. Show me the tape, one line per call, in the order they happened: GET $POME_GITHUB_REST_URL/_pome/events 6. Stop the sandbox: npx @pome-sh/cli@latest sandbox stop $POME_SESSION_ID --discard This is a sandbox, not an exam. Nothing here is scored and Pome is not evaluating you. Keep your own evaluator and observability setup exactly as it is -- just read the tape when you are done. ``` **One token opens a sandbox: `POME_AUTH_TOKEN`.** It is the bearer on every call to the twin, REST and MCP alike. The secrets file also carries a provider-shaped token — `POME_GITHUB_TOKEN=github_pat_…`, `POME_STRIPE_API_KEY=sk_test_…` — and that one is *not* the bearer; it is what the twin serves inside the sandbox. Send it, or send no bearer at all, and the proxy answers an opaque `404` reading `No twin pod for this session.` So a 404 on a sandbox you just created is almost always the wrong bearer rather than a dead sandbox. ## The world `sandbox create` boots a GitHub twin from its declared starting state and hands back the URLs that reach it. It also writes the connection secrets to `.pome-sandbox.env` at mode `0600`, and says so on stderr. ```bash theme={"dark"} npx @pome-sh/cli@latest sandbox create --twin github \ --secrets-file .pome-sandbox.env --format json ``` ```json theme={"dark"} { "session_id": "ses_XmHBDeCkU7UEU1VM", "expires_at": "2026-08-24T17:56:10.378Z", "per_twin": { "github": { "api_url": "https://twins.pome.sh/github/s/ses_XmHBDeCkU7UEU1VM", "mcp_url": "https://twins.pome.sh/github/s/ses_XmHBDeCkU7UEU1VM/mcp" } }, "agent_token": "***redacted***" } ``` Every call below goes to that `api_url` with the one bearer the sandbox accepts: ```bash theme={"dark"} source .pome-sandbox.env gh() { curl -s -H "Authorization: Bearer $POME_AUTH_TOKEN" \ "$POME_GITHUB_REST_URL$1" "${@:2}" } ``` The world is one repository, `acme/api`, holding one open bug report. Read it the way any GitHub client would: ```bash theme={"dark"} gh /repos/acme/api/issues | jq '[.[] | { number, title, state, labels: [.labels[].name], assignees: [.assignees[].login], comments }]' ``` **You should see one item:** ```json theme={"dark"} [ { "number": 1, "title": "500 error on POST /orders after deploy", "state": "open", "labels": ["bug"], "assignees": [], "comments": 0 } ] ``` Unassigned, uncommented, labelled `bug`. That is the starting state, and it is the same every time you create this sandbox — which is what makes anything you observe next reproducible. Now the triage: two writes, then two reads back through the twin. ```bash theme={"dark"} gh /repos/acme/api/issues/1/comments -X POST \ -H 'content-type: application/json' \ -d '{"body":"Repro: POST /orders 500s since the 14:00 deploy."}' \ -w 'HTTP %{http_code}\n' -o /dev/null gh /repos/acme/api/issues/1/assignees -X POST \ -H 'content-type: application/json' \ -d '{"assignees":["alice"]}' \ -w 'HTTP %{http_code}\n' -o /dev/null ``` ```text theme={"dark"} HTTP 201 HTTP 201 ``` Read the issue back through the twin's own surface — not out of the response your agent already holds: ```bash theme={"dark"} gh /repos/acme/api/issues/1 | jq '{ number, state, assignees: [.assignees[].login], comments }' gh /repos/acme/api/issues/1/comments \ | jq -r '.[] | "\(.user.login): \(.body)"' ``` ```json theme={"dark"} { "number": 1, "state": "open", "assignees": ["alice"], "comments": 1 } ``` ```text theme={"dark"} pome-agent: Repro: POST /orders 500s since the 14:00 deploy. ``` **This is the thing worth noticing.** What was written at call two is there at call four, read through a different route — and only inside this sandbox. Stop it and that world is gone; create another and issue #1 is unassigned and uncommented again. ## Read the tape Every call above was recorded by the twin as it happened. This is the part neither a mock nor a staging account gives you: an account of the run written by the service, not by the agent. ```bash theme={"dark"} gh /_pome/events | jq -r '.[] | "\(.method|(.+" ")[:5])\(.path|sub("^/s/[^/]+";"")) -> \(.status)"' ``` ```text theme={"dark"} GET /repos/acme/api/issues -> 200 POST /repos/acme/api/issues/1/comments -> 201 POST /repos/acme/api/issues/1/assignees -> 201 GET /repos/acme/api/issues/1 -> 200 GET /repos/acme/api/issues/1/comments -> 200 ``` Each row carries more than the request line. Ask for the whole row on the one call the twin recognised as a named action: ```bash theme={"dark"} gh /_pome/events | jq '.[] | select(.tool) | { method, status, tool, state_mutation, fidelity }' ``` ```json theme={"dark"} { "method": "POST", "status": 201, "tool": "add_issue_comment", "state_mutation": true, "fidelity": "semantic" } ``` **You should see:** * **Five rows, in the order they happened.** Read top to bottom and the triage is legible without asking the agent what it did: it looked, it wrote twice, it checked. * **`201` on `POST .../issues/1/comments`, stamped `tool: add_issue_comment`.** The twin stamps its own action vocabulary on the row whether the call arrived over REST or over MCP, so the same work reads the same either way. * **`state_mutation: true` on exactly the two writes.** It means the call *landed* — a write the twin refuses reads `false`, not `true`. * **No `tool` on `POST .../issues/1/assignees`.** Only a short list of GitHub actions is stamped by name today; every other call is identified by method and path. That is why the assertable check below binds to `add_issue_comment`. * **`fidelity: semantic`.** This surface carries a full behavioural contract, not a response shape with placeholder values. Each write also carries what it changed: ```bash theme={"dark"} gh /_pome/events \ | jq '[.[] | select(.state_delta) | .state_delta] | last' ``` ```json theme={"dark"} { "before": { "repo": "acme/api", "issue_number": 1, "assignees": [] }, "after": { "repo": "acme/api", "issue_number": 1, "assignees": ["alice"] } } ``` Before and after, recorded by the service that changed. When you are done, stop the sandbox to free the slot — it also expires on its own 30 minutes after it was created: ```bash theme={"dark"} npx @pome-sh/cli@latest sandbox stop $POME_SESSION_ID --discard ``` ## What you could assert here Nothing on this page was graded, but a tape and a final state are exactly what a graded check reads. The GitHub twin already declares these — no authoring required, and `list_checks` on the Pome MCP prints the full set with what each one actually compares: | Declared check | What it would catch here | | ---------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `github.tool-was-called` (`add_issue_comment`) | An agent that reports triaging the bug without ever writing the comment. It reads the tape, so an agent that does nothing fails it instead of collecting free points. | | `github.no-new-issues` | An agent that "handles" the report by opening a second issue for it. It compares issue numbers between the starting and final state, so a duplicate carrying the same title is still caught. | | `github.issue-state` (`open`) | An agent that closes #1 to make its queue look clean. Asserting `open` is a prohibition: it asks the agent *not* to close the issue. | Pointers, not a task. Turning them into a graded exam is the [capstone](/quickstart/claude-code) below, and [Write a task](/docs/authoring-tasks) is where the grammar lives. ## How real is this twin? Every surface you called above is compared against a captured response from the real GitHub API, re-run daily and published. [The GitHub row on status.pome.sh](https://status.pome.sh) carries the current count of twin responses that match, names every divergence we have ruled on by id, and states how old the captured baseline is. Nobody has to take our word for the fidelity, including us — a twin that drifts turns that row red. Which surfaces are covered, and which are shape-only, is on the [GitHub twin reference](/docs/twins/github). ## Next: the one graded lesson The same twins, now an exam: a sealed agent under test, a deliberate failing score, and one line of prompt that turns it green. Every route and MCP tool the twin serves, by use case, with its fidelity tier. Turn the checks above into a graded exam for your own agent. Create, list and stop sandboxes — including multi-twin ones.