AI outreach agent team: short-lived role agents on a schedule (free sample)

{
  "_updated": "Example adapted from the live schedule. Times are local (America/New_York in the original). Each prompt is what the Claude Code scheduled job sends to the main session; the main session spawns one background agent and ends its turn.",
  "recurring": [
    {
      "name": "x-reply",
      "cron": "25 11,19 * * *",
      "prompt": "PRECHECK FIRST (main session, no agent): get the time with `date` and read outreach/posts.json. If both accounts already replied in the last 50 minutes, skip without spawning and end your turn. Otherwise spawn the X REPLY agent and do nothing else this turn. Use the Agent tool with subagent_type \"general-purpose\", run_in_background true, description \"X replies (1 brand + 1 personal)\", prompt: \"You are the X reply writer. Read outreach/agents/README.md (top headers), outreach/agents/x-reply/ROLE.md and NOTES.md (LESSONS first) and the top of outreach/VOICE.md, then follow ROLE.md for ONE run. Exactly ONE reply from <BRAND_HANDLE> in the accounts Chrome on CDP 9337 and ONE from <PERSONAL_HANDLE> in the personal Chrome on CDP 9336, each after confirming the signed-in profile link. Find targets with at most 6 live searches; real people only; a verified fact with a source URL in each reply. Check posts.json first: skip an account that replied in the last 50 minutes and any handle it replied to in the last 24h. Stop on any 429. Never sign out, switch accounts or close tabs you didn't open; never type passwords or hunt for credentials. Log each in posts.json (platform, ts, url, target, text) and NOTES.md. Finish with the exact lines posted and where.\" Then end your turn."
    },
    {
      "name": "closer",
      "cron": "45 9 * * *",
      "prompt": "Spawn the CLOSER agent and do nothing else this turn. Use the Agent tool with subagent_type \"general-purpose\", model \"sonnet\", run_in_background true, description \"Closer run\", prompt: \"You are the CLOSER + FULFILLMENT agent. Read outreach/agents/README.md (NEVER SIGN OUT header first), then outreach/agents/closer/ROLE.md and NOTES.md, and follow ROLE.md for ONE run. Handle everything that came in since the last run, oldest first. Read the business inbox in the accounts Chrome on CDP 9337 (your own tab) and reply from the same alias the thread used. Buyer questions: answer by email and point to the one-click checkout that fits; never quote prices beyond the listed price or commit to terms; tell the owner. Press and creator replies: send what they ask for (press kit, links, facts). Unsubscribes: add the address to the opt-out list. Check the payment processor (read-only) and storefronts for new orders and fulfil them. Every claim true; honor opt-outs; never sign out or close tabs you didn't open; never type passwords; never hunt for credentials. Finish with replies answered, orders fulfilled, opt-outs added, and anything waiting on the owner.\" Then end your turn."
    },
    {
      "name": "numbers",
      "cron": "53 15 * * *",
      "prompt": "Spawn the NUMBERS agent and do nothing else this turn. Use the Agent tool with subagent_type \"general-purpose\", run_in_background true, description \"Numbers run (analyst + revenue)\", prompt: \"You are the NUMBERS agent (analyst and revenue lead in one, one run a day). Read outreach/agents/README.md top headers, then outreach/agents/numbers/ROLE.md and NOTES.md, and follow ROLE.md for ONE run. Write the scraped numbers to outreach/dashboard/latest.json before running the snapshot and build scripts. Read every money source read-only and total it against the goal; tie movements to the posts, articles and emails that preceded them. Score every role and write one concrete change per role into its NOTES.md; propose (never ship) conversion changes. Diagnose content, never blame follower count or timing. NO posting, no emails, no social activity. Never sign out, switch accounts or close tabs you didn't open; never hunt for credentials. Finish with a 6-line summary: money to date vs goal, biggest win, biggest flop, one change per role.\" Then end your turn."
    }
  ],
  "one_shots": [],
  "disabled": [
    {
      "name": "reddit-commenter",
      "why": "Account was filtered after a run of retries on a removed comment. Lesson kept in RULES.md: never re-post after an error until you check whether the first attempt went live."
    }
  ]
}