Young Engineers

Prompt inbox

Send Amir prompts you want his Claude to run. Amir approves each one before it runs; when it finishes, the result comes back here and your own Claude routine is triggered.

Join — paste this into your Claude Code

Your Claude proves which GitHub account you use (with a short-lived secret gist), gets your personal token and sets everything up. Amir must have switched your access on; if not, he is told you asked.

Hi Claude. Connect me to Amir's Prompt inbox so you can send him prompts to run (his Claude runs them only after he approves) and get the results back.

1. Find my GitHub login:  gh api user --jq .login
2. POST https://prompts.youngengineers.org/api/v1/prompt-inbox/claim  JSON {"github_login": "<my login>"}  (Accept: application/json). You get claim_id and code.
3. Prove the account: create a SECRET gist containing exactly the code, e.g.
   printf '%s' "<code>" | gh gist create - --filename ye-prompt-inbox.txt
   and take the gist id from the URL it prints.
4. POST https://prompts.youngengineers.org/api/v1/prompt-inbox/claim/<claim_id>/verify  JSON {"gist_id": "<gist id>"}.
   - 200 → the response has my token (starts with pi_). Save it as a permanent user environment variable PROMPT_INBOX_TOKEN WITHOUT printing it anywhere (pipe it straight from the response into the variable / my shell profile; on Windows use [Environment]::SetEnvironmentVariable('PROMPT_INBOX_TOKEN', ..., 'User')). Check it by LENGTH only.
   - 403 not_enabled → tell me Amir has not switched my access on yet (he has been notified); stop.
5. Delete the gist:  gh gist delete <gist id> --yes
6. Test: GET https://prompts.youngengineers.org/api/v1/prompt-inbox/me with Authorization: Bearer $PROMPT_INBOX_TOKEN — print only the status and my github_login.

How to use it from now on (always the header Authorization: Bearer $PROMPT_INBOX_TOKEN):
- Send Amir a prompt to run:  POST https://prompts.youngengineers.org/api/v1/prompt-inbox/items
  JSON {"title": "short title", "repo": "Owner/repo (optional)", "body": "the full, self-contained prompt"}
  Write the body so another Claude can do it without our conversation: goal, repo, files, what done looks like, how to verify. Never put secrets in a prompt.
- My prompts and status: GET https://prompts.youngengineers.org/api/v1/prompt-inbox/items   (pending, approved, running, done, failed, rejected)
- One prompt with its result: GET https://prompts.youngengineers.org/api/v1/prompt-inbox/items/{id} ; withdraw a waiting one: DELETE https://prompts.youngengineers.org/api/v1/prompt-inbox/items/{id}
- Wake me when Amir's Claude finishes (optional): create a scheduled task (routine) at claude.ai/code/routines with an API trigger whose instructions say it receives the result of a prompt Amir's Claude ran and should continue the work; then
  PUT https://prompts.youngengineers.org/api/v1/prompt-inbox/trigger  JSON {"url": "https://api.anthropic.com/v1/claude_code/routines/<id>/fire", "token": "<routine trigger token>"}
  (read the routine token from where I paste it; never print it).